Data processing system

Abstract

The present application relates to a unified ID management system including a user information management unit, a permission assignment management unit, an audit record management unit, a generative AI processing unit, and the like, for the purpose of efficientization and security enhancement of account and permission management of a large number of business entrusting employees in a communication industry or IT-related business, using generative AI to analyze the attributes, business content, past permission assignment performance, and the like of each employee, automatically formulating and applying a minimum permission assignment policy. In addition, it is characterized in that the permission is automatically assigned, changed, and deprived according to events such as employment, transfer, and resignation, the audit records are collected in detail and abnormality detection is performed, synchronization between systems is performed through API linkage, and single sign-on is realized.

Description

Technical Field

[0001] The technology disclosed herein relates to a system. Background Technology

[0002] Patent Document 1 discloses a personalized chatbot control method executed by at least one processor, the method comprising: receiving user speech; adding the user speech to a prompt containing instructions related to a chatbot role; encoding the prompt; and inputting the encoded prompt into a language model to generate chatbot speech in response to the user speech.

[0003] Patent document 1: Japanese Patent Application Publication No. 2022-180282.

[0004] The problem this invention aims to solve is the increasing complexity of account management in telecommunications or IT-related companies when receiving a large number of business assignments. This includes the increased management burden due to issuing and managing system permissions individually for each employee; increased risks of information leakage and unauthorized access due to incorrect permission settings; insufficient accumulation of proprietary technology within the company due to the personalization of system management tasks; increased costs of account issuance and management; and time lags until business commencement due to procedural delays. To address these problems, the invention aims to improve account management by introducing and utilizing proprietary technology. Chengshi AI's unified ID management system unifies account and permission management across multiple systems. Through unified directory services and integration with the HR system, it enables automatic updates of accounts and permissions during onboarding, transfers, and departures. It automatically applies minimum permission assignment strategies defined by department and position, achieves efficient inter-system communication through API integration, and ensures consistent authentication management in hybrid environments through the unification of cloud infrastructure and local environments. It also facilitates the recording and tracking of access history through audit logs, eliminates personalization, and accumulates proprietary technologies, thereby simultaneously achieving cost reduction and business efficiency. Summary of the Invention

[0005] As a means to solve the problem, a system is constructed, including a user information management department, a permission granting management department, an audit record management department, and a generative AI processing department. In the user information management department, the individual attribute information, department, position, and employment status of business employees engaged in telecommunications or IT-related businesses are uniformly managed. In the permission granting management department, generative AI is used to automatically formulate the minimum permission granting strategy corresponding to each user's department and position, and it is linked with the onboarding, transfer, and resignation event information from the human resources system to execute the allocation, modification, and revocation of permissions in real time. In the audit record management department, all account operations, permission changes, system access history records, and time information are recorded together. In the generative AI processing department, the information from these departments is analyzed, and the signs of permission setting errors and unauthorized access are automatically detected. In addition, information is synchronized with the cloud infrastructure and the local environment's authentication infrastructure through API linkage, which achieves consistent ID management and reduces security risks in a hybrid environment. Attached Figure Description

[0006] Figure 1 This is a conceptual diagram illustrating an example of the configuration of a data processing system according to the first embodiment.

[0007] Figure 2 This is a conceptual diagram illustrating an example of the main functions of the data processing apparatus and smart device according to the first embodiment.

[0008] Figure 3 This is a conceptual diagram illustrating an example of the data processing system configuration in the second embodiment.

[0009] Figure 4 This is a conceptual diagram illustrating an example of the main functions of the data processing device and smart glasses according to the second embodiment.

[0010] Figure 5 This is a conceptual diagram illustrating an example of the data processing system configuration in the third embodiment.

[0011] Figure 6 This is a conceptual diagram illustrating an example of the main functions of the data processing device and head-mounted terminal according to the third embodiment.

[0012] Figure 7 This is a conceptual diagram illustrating an example of the data processing system configuration in the fourth embodiment.

[0013] Figure 8 This is a conceptual diagram illustrating an example of the functions of the main parts of the data processing device and robot according to the fourth embodiment.

[0014] Figure 9 It represents an emotion graph that maps multiple emotions.

[0015] Figure 10It represents an emotion graph that maps multiple emotions.

[0016] Explanation of reference numerals in the attached figures Data processing systems 10, 210, 310, and 410 12 Data processing device 14 Smart devices 214 Smart Glasses 314 Head-mounted terminal 414 Robot. Detailed Implementation

[0017] Hereinafter, an example of an implementation of the system involved in this disclosure will be described with reference to the accompanying drawings.

[0018] First, let's explain the terms used in the following description.

[0019] In the following embodiments, the processor (hereinafter referred to as "processor") can be a single computing device or a combination of multiple computing devices. Furthermore, a processor can be a single computing device or a combination of multiple computing devices. Examples of computing devices include CPU (Central Processing Unit), GPU (Graphics Processing Unit), GPGPU (General-Purpose computing on Graphics Processing Units), APU (Accelerated Processing Unit), etc.

[0020] In the following implementation, the labeled RAM (Random Access Memory) is a memory that temporarily stores information and is used by the processor as working memory.

[0021] In the following embodiments, the labeled memory is one or more non-volatile storage devices used to store various programs and various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), disk (e.g., hard disk) or magnetic tape, etc.

[0022] In the following implementation, the labeled Communication I / F (Interface) is an interface that includes a communication processor and an antenna, etc. The Communication I / F is responsible for communication between multiple computers. Examples of communication standards applicable to the Communication I / F include wireless communication standards such as 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

[0023] In the following implementation, "A and / or B" is synonymous with "at least one of A and B". That is, "A and / or B" means that it can be only A, only B, or a combination of A and B. Furthermore, in this specification, when "and / or" connects more than three items, the same approach as "A and / or B" applies.

[0024] [First Implementation] Figure 1 An example of the configuration of the data processing system 10 according to the first embodiment is shown.

[0025] like Figure 1 As shown, the data processing system 10 includes a data processing device 12 and an intelligent device 14. An example of the data processing device 12 is a server.

[0026] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0027] The smart device 14 includes a computer 36, a receiver 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the receiver 38, output device 40, camera 42, and communication I / F 44 are also connected to the bus 52.

[0028] The receiving device 38 includes a touchscreen 38A and a microphone 38B, etc., for receiving user input. The touchscreen 38A receives user input generated by contact with an indicator (e.g., a pen or finger). The microphone 38B receives user input generated by sound by detecting the user's voice. The control unit 46A in the processor 46 sends data representing user input received through the touchscreen 38A and microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data representing the user input.

[0029] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting data in a form perceptible to the user 20 (e.g., sound and / or text). The display 40A displays visual information such as text and images according to instructions from the processor 46. The speaker 40B outputs sound according to instructions from the processor 46. The camera 42 is a small digital camera equipped with an optical system such as a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0030] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for sending and receiving various information between processor 46 and processor 28 via network 54.

[0031] Figure 2 An example of the main functions of the data processing device 12 and the smart device 14 is shown.

[0032] like Figure 2 As shown, in the data processing apparatus 12, specific processing is performed by the processor 28. A specific processing program 56 is stored in the memory 32. The specific processing program 56 is an example of a "program" as understood in this disclosure. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 based on the specific processing program 56 executed on the RAM 30.

[0033] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0034] In the smart device 14, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The receive / output program 60 is used in conjunction with the data processing system 10 and the specific processing program 56. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart device 14 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290. The receive / output processing is implemented by the processor 46 acting as a control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0035] In addition, other devices besides the data processing device 12 may also have a data generation model 58.

[0036] For example, a server device (e.g., a generation server) may have a data generation model 58.

[0037] In this case, the data processing device 12 obtains the processing results (prediction results, etc.) using the data generation model 58 by communicating with a server device that has the data generation model 58. Furthermore, the data processing device 12 can be a server device or a user-owned terminal device (e.g., a mobile phone, robot, home appliance, etc.). Next, an example of the processing performed by the data processing system 10 of the first embodiment will be described.

[0038] (Example 1) The specific processing flow in Example 1 will be described below. Each part of the system described below is implemented by the data processing device 12 and the smart device 14. Furthermore, the data processing device 12 is referred to as the "server" and the smart device 14 is referred to as the "terminal".

[0039] (Implementation methods for carrying out the present invention) As an embodiment of this invention, a structure connecting a server and a terminal via a network is adopted. On the server side, a user information management department, a permission granting management department, an audit log management department, and a generative AI processing department are installed. On the terminal side, a user interface department and a portion of the authentication processing department are installed. The structure and function of each department, the specific application method of generative AI, and the division of responsibilities between the server and the terminal are explained in detail below with specific examples. Specifically, this system functions as a unified ID management infrastructure for tens of thousands of employees in a large-scale multinational corporation. As a hardware structure, this system possesses a parallel computing cluster using image processing semiconductors. The server combines multiple computing nodes through a high-speed network interface. The server deploys a distributed in-memory database. The server containerizes and executes each management department as a microservice architecture. The terminal executes a web browser or a dedicated client application. The terminal communicates with the server via a secure encrypted communication protocol. This system distributes the processing load across multiple server nodes. This system reduces communication latency. This system eliminates single points of failure. This system achieves high availability and scalability. This system completely eliminates the manual permission management tasks previously performed by humans. This system dramatically improves the processing efficiency of computer technology itself.

[0040] First, the User Information Management Department runs on a server and centrally manages a wide variety of attribute information for employees who have commissioned business, including their name, employee ID, department, position, employment status (current, resigned, transferred, etc.), contact information, business start date, scheduled business end date, responsible projects, scheduled system usage, past transfer history, skill set, qualification information, and emergency contact information. For example, to efficiently manage information for over 22,000 employees, distributed databases, clustering technology, and database sharding technology are used on the server side. Furthermore, the User Information Management Department integrates with the HR system and external contract management systems via API, receiving real-time information on employee onboarding, transfers, resignations, contract updates, and contract terminations, and automatically updating user information. For instance, when an employee is assigned to a new project, they receive an API notification from the project management system, instantly reflecting their project information. Specifically, this User Information Management Department uses a document-oriented, non-relational database. This database stores each user's attribute information as a hierarchical key-value data structure. The User Information Management Department uses a stateless communication architecture when communicating with external systems. This User Information Management Department receives payloads in data exchange format from external systems. It then performs syntax parsing on the received payloads. The extracted attribute data is converted into multi-dimensional numerical vectors. These converted numerical vectors are indexed and stored in a vector database. This process dramatically improves data retrieval speed. It ensures data consistency between systems in milliseconds. This system is extremely effective in the development of information and communication technologies where project structures change frequently. It reduces data management overhead and improves the utilization efficiency of computer storage areas.

[0041] The permission management system runs on the server, utilizing generative AI to comprehensively analyze each user's department and position, business content, past permission granting performance, system usage history, project information, skill sets, and qualification information to automatically formulate a minimum permission granting strategy. For example, it grants development department managers "management permissions for the source code management system," "access permissions for the development environment," and "full access permissions for project management tools," while granting CS department employees only "viewing permissions for the customer information management system" and "input permissions for the consultation response system." The AI ​​automatically determines the optimal permission set based on department, position, and business content. Furthermore, the AI ​​continuously learns from past permission granting history and access logs, automatically extracting exception patterns such as "even for the same position, additional permissions are needed when assigned to a specific project," and flexibly updating permission granting rules. Through API integration with the HR system, it receives real-time information on employee onboarding, transfers, departures, contract updates, and contract terminations, instantly allocating, changing, or revoking permissions for corresponding users. For example, in the case of a transfer from the Development Department to the CS Department, the AI ​​detects the transfer event and automatically revokes the "administrative privileges of the source code management system" granted in the Development Department, only granting new "viewing privileges of the customer information management system" in the CS Department. Upon resignation or contract termination, the AI ​​automatically deletes the corresponding user's account from all relevant systems, revoking all privileges. Furthermore, the AI ​​also handles complex situations such as "the same user having privileges across multiple departments" or "temporary privileges granted for temporary participation in projects," granting privileges only for the required period and automatically revoking them after the period ends. Specifically, this privilege granting management department installs a neural network based on a large-scale language model architecture. The input data to this neural network is a 768-dimensional numerical vector representing the user's attribute information. This input data includes department identifiers, job identifiers, and skill vectors as elements. This neural network performs matrix operations on the input data using a self-attention mechanism. The output data from this neural network is a one-dimensional numerical array with continuous values ​​from 0.0 to 1.0 representing the probability of granting access permission to each system. The first element of this output data represents the probability of granting administrative privileges to the source code management system. The second element of this output data represents the probability of granting viewing privileges to the customer information management system. This permission grants the management department the authority to determine whether the output probability value exceeds a threshold of 0.8. If the threshold is exceeded, the management department is authorized to validate the corresponding permission assignment flag. This neural network continuously updates the weight parameters using backpropagation of the cross-entropy error function. This process mathematically prevents excessive permission assignment. This process reduces the risk of information leakage and improves the overall security level of the system. This system implements non-intuitive, rule-based processing that does not rely on human intuition.

[0042] The audit log management system runs on a server, meticulously recording all account operations (new registration, information updates, deletions), permission changes (granting, modifying, revoking), system access history (login, logout, operation content), data synchronization history with external systems via API, terminal information (IP address, terminal identifier, browser information, etc.), and time information. For example, when a user is newly registered, the system records in detail which terminal, which administrator, what permissions were granted, and which system the account was issued to, enabling future retrieval or investigation. Furthermore, to prevent tampering with audit logs, the audit log management system can utilize blockchain or electronic signature technology. For instance, for important permission changes or account deletion operations, multi-factor authentication by the administrator is required, and the operation record is electronically signed to prevent subsequent tampering or denial. Specifically, this audit log management system constructs a hash chain structure based on distributed ledger technology in memory. The system inputs each operation log data into a cryptographic hash function. The system outputs a fixed-length 256-bit hash value. This audit log management department combines the hash value of the immediately preceding log data into the current log data. It then calculates a new hash value from the combined data. Next, it generates a digital signature based on the administrator's private key using elliptic curve cryptography. This digital signature is then assigned to the log data. The log data is stored in a time-series database. An inverted index is built to optimize the execution speed of retrieval queries. This process mathematically guarantees the integrity of the log data. Physically, it makes subsequent data tampering impossible. This system is extremely effective in environments with stringent compliance requirements, such as financial institutions or medical facilities. This system fully automates audit response processes, reducing the processing load on computers.

[0043] The generative AI processing unit runs on a server, comprehensively analyzing diverse data such as user information, permission granting history, audit logs, system utilization, and event information from external systems. For example, the AI ​​learns from past permission granting performance and business content, system utilization frequency, and continuously updates the optimal permission granting strategy. Using anomaly detection algorithms, the AI ​​detects unusual access and permission changes in real time, automatically extracting anomaly patterns such as "users who don't usually access important systems late at night," "the same user repeatedly granting permissions in multiple systems within a short period," and "the account of a former user being illegally used," and notifying administrators of these risks. The AI ​​analyzes administrator operation history and past audit logs to identify errors in permission settings and tendencies towards personalization, automatically reporting risks such as "frequent permission granting errors caused by specific administrators" and "permission settings deviating from standards in specific departments," and can also propose corrective measures and educational initiatives. Specifically, this generative AI processing unit uses an anomaly detection model based on a Long Short-Term Memory (LSTM) network. The input data to this anomaly detection model is a 128-dimensional time-series tensor containing access frequency and geographic distance scores of Internet Protocol (IP) addresses over the past 24 hours as elements. This anomaly detection model computes a hidden state vector from the input temporal tensor. The output data from this anomaly detection model consists of anomaly scores represented by continuous values ​​from 0.0 to 1.0. The generative AI processing unit determines whether the output anomaly score exceeds a threshold of 0.85. If the threshold is exceeded, the generative AI processing unit generates a control signal to immediately lock the corresponding user's account. The generative AI processing unit sends the control signal to the authentication infrastructure. The generative AI processing unit pushes a notification warning message to the administrator's terminal. This process blocks internal misconduct in a zero-trust network environment in milliseconds. This process achieves pattern recognition in high-dimensional spaces that exceed the limits of human surveillance capabilities. This system dramatically reduces the probability of security incidents.

[0044] Furthermore, generative AI synchronizes information with cloud infrastructure and local environment authentication infrastructure through API linkage. For example, for diverse systems such as cloud storage services, groupware, business applications, virtual desktops, email systems, file servers, and network device management systems, AI automatically parses the API specifications and authentication protocols of each system, appropriately converting and synchronizing the required authentication and permission information. For instance, if a new folder is created in a cloud storage service, AI detects the event and automatically grants access permissions only to employees in the relevant department. Or, if a business-delegated employee leaves, AI automatically deletes the corresponding user's account from all relevant systems, recording a complete deletion history as an audit record. Specifically, this generative AI processing unit uses a natural language processing model to automatically parse the specification documents of external systems. The input data to this natural language processing model is a text tensor describing the interface specifications of the external system. This natural language processing model extracts the Uniform Resource Locator (URL), required parameters, and authentication methods of the endpoints from the text tensor. The output data from this natural language processing model is a data object that structures the extracted information. This generative AI processing unit dynamically generates Hypertext Transfer Protocol (HTTP) request messages based on the output data object. This generative AI processing unit sends the generated request messages to the target system. This processing completely eliminates the manual programming time required for system integration. This processing prevents communication errors caused by misconfiguration. This system minimizes data synchronization latency in heterogeneous environments. This system dramatically improves the overall efficiency of computer networks.

[0045] A user interface unit is installed on the terminal side, providing a screen for employees and administrators to confirm their account information and permission status, and to submit and receive approvals. For example, when an employee applies to use a new business system, the application information is sent from the terminal to the server. The server-side generative AI analyzes the application content, automatically determines whether the required permissions are appropriate, and returns approval or rejection feedback to the terminal. Administrators can view audit logs and AI-based risk alerts on the terminal and instruct additional investigations and corrective actions as needed. Furthermore, an authentication processing unit is installed on the terminal side, combining multi-factor authentication, biometric authentication, and terminal authentication to prevent impersonation and unauthorized access. For example, when an administrator makes important permission changes, a one-time password or fingerprint authentication is required on the terminal, and the authentication result is sent to the server to authorize the operation. Specifically, the generative AI processing unit on the server side has an inference engine to determine the appropriateness of the application content. The input data to this inference engine consists of natural language text describing the application reason and a numerical vector representing the user's current permission status. This inference engine calculates the semantic similarity between the input data. The output data from this inference engine consists of a text string representing the approval recommendation score (an integer value from 0 to 100) and the rejection reason. This server performs automatic approval processing when the approval recommendation score is 90 or higher. This server performs automatic rejection processing when the approval recommendation score is less than 50. The authentication processing unit on this terminal side processes the image tensor obtained from the fingerprint sensor. This authentication processing unit extracts feature points from the image tensor and calculates hash values. This authentication processing unit sends the calculated hash values ​​to this server. This process avoids the direct transmission of biometric images, which consumes network bandwidth. This process eliminates bottlenecks in the approval process, minimizing service downtime. This system reduces the cognitive load on administrators and improves computer responsiveness.

[0046] Furthermore, AI can analyze past audit logs to identify frequent errors in permission settings within specific departments and positions, suggesting to administrators the necessity for policy reassessment and training. AI continuously learns about changes in the business content and system usage of delegated employees. For example, when a new business system is introduced, it automatically generates optimal permission assignment rules, significantly reducing the administrator's workload. Moreover, AI can automatically control permissions, such as granting access to systems accessible only to employees with specific qualifications, based on the skill sets and qualification information of delegated employees. Specifically, this generative AI processing unit uses clustering algorithms to extract patterns in a high-dimensional data space. The input data to this clustering algorithm is a multi-dimensional matrix combining past permission change history and event history. This generative AI processing unit calculates the Euclidean distance between data points and forms clusters. The output data from this clustering algorithm is the cluster centroid vector representing high-risk permission setting patterns. This generative AI processing unit calculates the similarity between the vector of a new permission setting request and the cluster centroid vector. This generative AI processing unit automatically blocks the setting request if the similarity exceeds a predetermined threshold. This generative AI processing unit performs optical character recognition (OCR) processing on the image data of qualification certificates. It uses a named entity recognition model to determine the qualification name and validity period from the extracted text. This processing implements proactive risk management as an autonomous action of the computer. This processing prevents compliance violations before they occur. This system eliminates the need for human visual verification, dramatically increasing processing speed.

[0047] Furthermore, for example, AI can uniformly manage authentication information across multiple cloud services and local systems, enabling users to access various systems from their terminals via single sign-on. For instance, after an employee authenticates once on a terminal, AI automatically generates authentication tokens for each system, allowing users to seamlessly access business applications, cloud storage, email systems, etc., without requiring additional authentication. AI monitors the authentication status and usage of each system in real time, immediately locking the corresponding account and notifying the administrator of alerts upon detecting abnormal authentication requests or unauthorized token usage. Specifically, this generative AI processing unit dynamically generates authentication assertions based on a security assertion markup language. This generative AI processing unit possesses an inference model that dynamically adjusts the token's validity period. The input data to this inference model is a context vector containing the user's current Internet Protocol address, the device's security status, and the access time. The output data from this inference model is the token's validity period, expressed as a value in seconds. This generative AI processing unit sets the validity period to 300 seconds when the risk score is deemed high. This generative AI processing unit sets the validity period to 3600 seconds when the risk score is deemed low. This generative AI processing unit digitally signs the generated token and sends it to this terminal. This process mathematically optimizes the trade-off between convenience and security. This process minimizes the probability of successful unauthorized session hijacking attacks. This system reduces user authentication waiting time and improves overall system availability.

[0048] In this embodiment of the invention, by centrally managing the user information management department, permission granting management department, audit record management department, and generative AI processing department on the server side, and installing a user interface department and a portion of the authentication processing department on the terminal side, the challenges of large-scale ID management in the telecommunications and IT-related enterprises can be comprehensively and efficiently addressed. This includes automating the management of employee accounts and permissions, reducing security risks, lowering operating costs, eliminating the need for personal identification, accumulating proprietary technologies, enhancing audit response capabilities, improving the efficiency of inter-system collaboration, implementing single sign-on, providing access control based on qualification information, and strengthening security based on multi-factor authentication. Generative AI plays a core role in the overall system, such as automatically formulating permission granting policies, detecting anomalies, synchronizing systems based on API collaboration, parsing audit records, automatically notifying of risk alerts, proposing policy reassessment, single sign-on control, and managing qualification information, thus becoming an indispensable element in realizing the invention. Specifically, the system performs asynchronous message queue communication between modules. The system employs an integrated learning architecture that combines multiple artificial intelligence models. The system weights and sums the inference results of each model and issues the final judgment. This system features an orchestration engine that dynamically optimizes the allocation of computing resources. It automatically scales container replicas horizontally based on traffic surges. It processes millions of access requests from globally deployed enterprise-level systems without latency. This system transcends mere business efficiency, addressing the technical challenges of improving the efficiency of traffic control and memory management in computer networks. It overcomes the physical limitations of hardware through advanced software algorithms. This system functions as a next-generation autonomous information security infrastructure.

[0049] (System Composition) The system involved in this implementation includes a user information management department, a permission granting management department, an audit record management department, a generative AI processing department, a user interface department, an authentication processing department, and an API linkage department. The user information management department centrally manages a variety of attribute information in the database for employees engaged in telecommunications or IT-related business assignments, including their names, employee numbers, departments, positions, employment status, contact information, work start dates, scheduled work end dates, responsible projects, scheduled system usage, past transfer history, skill sets, qualification information, and emergency contact information. For example, to efficiently manage information on over 22,000 employees, distributed databases, cluster technology, and database sharding technology can be used. Furthermore, the user information management department links with the human resources system and an external contract management system via API, receiving real-time information on employee onboarding, transfers, departures, contract updates, and contract terminations, and automatically updating user information. For example, when an employee is assigned to a new project, they receive an API notification from the project management system, instantly reflecting the project information. Furthermore, the user information management department can automatically determine whether specific business and system utilization conditions are met based on employees' skill sets and qualification information. For example, in cases where only employees with information security qualifications can access a specific system, the system automatically compares the qualification information to determine whether access is permitted. Specifically, this system configures each processing unit in an independent memory address space. The user information management department constructs a B-tree index structure to improve database retrieval performance. The user information management department has a dedicated inference module for determining the suitability of qualification information. The input data to this inference module consists of text data of the qualification requirements specified by the system and a numerical vector representing the qualifications held by the user. This inference module calculates the cosine similarity between the two sets of data. The output data from this inference module is a suitability score represented by continuous values ​​from 0.0 to 1.0. If the suitability score is above a threshold of 0.9, the user information management department writes an access permission flag into a register in memory. This process significantly reduces the execution time of retrieval queries with complex conditional branches. This process enables real-time access control. This system avoids database lock contention, maximizing transaction throughput.

[0050] The permission granting management department utilizes generative AI to comprehensively analyze each user's department and position, business content, past permission granting performance, system usage history, project information, skill sets, and qualification information to automatically formulate a minimum permission granting strategy. For example, the development department manager is granted "management permissions for the source code management system," "access permissions for the development environment," and "full access permissions for project management tools," while CS department employees are granted only "viewing permissions for the customer information management system" and "input permissions for the consultation response system." The AI ​​automatically determines the optimal permission set based on department, position, and business content. Furthermore, the AI ​​continuously learns from past permission granting history and access logs, automatically extracting exception patterns such as "even for the same position, additional permissions are needed when assigned to a specific project," and flexibly updating permission granting rules. Through API linkage from the HR system, real-time information on employee onboarding, transfers, departures, contract updates, and contract terminations is received, allowing for the immediate allocation, modification, or revocation of permissions for corresponding users. For example, in the case of a transfer from the Development Department to the CS Department, the AI ​​detects the transfer event and automatically revokes the "management privileges of the source code management system" granted in the Development Department, only granting new "viewing privileges of the customer information management system" in the CS Department. Upon resignation or contract termination, the AI ​​automatically deletes the corresponding user's account from all relevant systems, revoking all privileges. Furthermore, the AI ​​can handle complex situations such as "the same user having privileges across multiple departments" and "granting temporary privileges for temporary project participation," granting privileges only for the necessary period and automatically revoking them after the period ends. For example, it is also possible to temporarily grant system management privileges for short-term projects and automatically revoke them after the project ends. Specifically, this permission granting is implemented by an autonomous agent in the management department that uses reinforcement learning algorithms. The input data to this autonomous agent is a state vector representing the user's current set of privileges and assigned business tasks. The autonomous agent selects the action of adding or deleting privileges based on the state vector. The output data from this autonomous agent is an action value matrix representing the expected reward value of each action. This permission granting is implemented by the management department to select the action with the maximum value from the action value matrix and generate a permission change command. This autonomous intelligent agent receives positive rewards for successful task completion and negative rewards for security incidents. The agent updates the weights of its internal neural network based on these reward signals. This process enables the system to autonomously adapt to dynamic environmental changes. It resolves complex permission dependencies that static, rule-based systems cannot handle. This system fundamentally eliminates the security vulnerability of excessive permissions.

[0051] The audit log management department meticulously records all account operations (new registration, information updates, deletion), permission changes (granting, modifying, revoking), system access history (login, logout, operation content), data synchronization history with external systems via API, terminal information (IP address, terminal identifier, browser information, etc.), and time information. For example, when a user is newly registered, the department records in detail which terminal, which administrator, what permissions were granted, and which system the account was issued to, enabling future retrieval or investigation. Furthermore, to prevent tampering with audit logs, the audit log management department can utilize blockchain or electronic signature technology. For example, for important permission changes or account deletion operations, multi-factor authentication by the administrator is required, and the operation record is electronically signed to prevent subsequent tampering or denial. The audit log management department also has regular backup and archiving functions for long-term storage of audit logs. For example, it securely stores all permission change history records from the past 5 years, allowing for rapid retrieval and extraction as needed. Specifically, this audit log management department is equipped with a data compression algorithm for efficiently processing massive log data. This audit log management system uses dictionary-based compression techniques to reduce the size of log data. It encrypts the compressed data using Advanced Encryption Standard (AES) algorithms. It implements inter-node consensus algorithms in distributed ledger technology. It synchronizes log data replication across multiple server nodes. It automatically schedules data migration to archive storage. This process significantly reduces the consumption of physical storage capacity on storage devices. This process shortens data transfer times across networks. This system maintains millisecond-level retrieval response performance for petabyte-level audit data. This system provides improvements in computer technology that simultaneously enhance data availability and confidentiality.

[0052] The generative AI processing unit comprehensively analyzes diverse data, including user information, permission granting history, audit logs, system utilization, and event information from external systems. The AI ​​learns from past permission granting performance, business content, and system utilization frequency, continuously updating the optimal permission granting strategy. Using anomaly detection algorithms, the AI ​​detects unusual access and permission changes in real time, automatically extracting abnormal patterns such as "users who don't usually access important systems late at night," "the same user repeatedly granting permissions across multiple systems within a short period," and "the account of a former user being illegally used," and notifying administrators of these risks. From administrator operation history and past audit logs, the AI ​​analyzes permission setting errors and tendencies towards personalization, automatically reporting risks such as "frequent permission granting errors caused by specific administrators" and "permission settings deviating from standards in specific departments," and can also propose corrective measures and the necessity of training. Furthermore, the AI ​​synchronizes information with cloud infrastructure and local authentication infrastructure through API linkage. For example, for diverse systems such as cloud storage services, groupware, business applications, virtual desktops, email systems, file servers, and network device management systems, AI automatically parses the API specifications and authentication protocols of each system, appropriately converting and synchronizing the required authentication and permission information. AI monitors the authentication status and utilization of each system in real time, and immediately locks the corresponding account and notifies the administrator of alerts upon detecting abnormal authentication requests or unauthorized token use. AI can also automatically control permissions, such as granting access to systems accessible only to employees with specific qualifications, based on the skill sets and qualification information of employees entrusted with business tasks. Specifically, this generative AI processing unit employs a multimodal learning architecture. The input data to this multimodal learning architecture consists of structured data composed of numerical vectors from logs and unstructured data composed of text tensors from specification documents. This generative AI processing unit uses a cross-attention mechanism to calculate the correlation between heterogeneous data. The output data from this generative AI processing unit is a data object describing the anomaly probability score and synchronization settings. When the anomaly probability score exceeds a specified threshold, this generative AI processing unit expands and executes an automatic blocking script in memory. This process enables early detection of advanced network attacks that may not be detectable with a single data source. It significantly improves the system's defense against unknown threats. This system reduces the overhead of data conversion in complex inter-system interactions and optimizes processor efficiency.

[0053] The user interface provides a screen for employees and administrators to confirm their account information and permissions, and to submit and receive approvals. For example, when an employee applies to use a new business system, they send the application information from the terminal to the server. The server-side generative AI analyzes the application content, automatically determines whether the required permissions are appropriate, and returns feedback to the terminal indicating approval or rejection. Administrators can view audit logs and risk alerts generated by the AI ​​from the terminal and instruct additional investigations and corrective measures as needed. Furthermore, the user interface also provides an input screen for employees to update their skills and qualifications, and to submit applications for transfers and resignations. For example, if an employee obtains new qualifications, they can input this information from the terminal, and the AI ​​can automatically grant access to the corresponding system. Specifically, this user interface performs differential update rendering processing using a virtual document object model. This user interface establishes a real-time bidirectional communication session with the server using the WebSocket protocol. The generative AI processing unit on the server side has a predictive model to assist user input operations. The input data to this predictive model is the text string that the user is currently typing. The output data from this predictive model is a list of predicted text candidates for the application reasons. This user interface section displays the received text candidate list on the screen. This process avoids redrawing the entire screen, reducing the processor load on the terminal. This process reduces network traffic, saving bandwidth. This system proactively prevents user input errors, improving data quality.

[0054] The authentication processing unit implements personal authentication for users and administrators accessing the system. For example, it employs multi-factor authentication such as combined password authentication, one-time password authentication, biometric authentication, and terminal authentication to prevent impersonation and unauthorized access. When an administrator makes important permission changes, one-time password authentication and fingerprint authentication are required on the terminal, and the authentication result is sent to the server to authorize the operation. Furthermore, the authentication processing unit sends authentication history records to the audit record management department to record all authentication events. Specifically, this authentication processing unit implements a challenge-response authentication protocol based on public-key cryptography. This authentication processing unit also implements passwordless authentication. The biometric authentication module on the terminal side executes a convolutional neural network. The input data to this convolutional neural network is an image tensor of 256 pixels vertically and 256 pixels horizontally obtained from the fingerprint sensor. This convolutional neural network extracts local feature maps from the image tensor. The output data from this convolutional neural network is a continuous score from 0.0 to 1.0 representing the similarity to a registered template. This authentication processing unit erects an authentication success flag when the score is above 0.95. This authentication processing unit uses a private key to sign the authentication assertion. This process completely eliminates the need for passwords to be sent over the network, rendering man-in-the-middle attacks ineffective. It also reduces authentication latency and improves user experience.

[0055] The API linkage unit interacts with external systems such as HR, contract management, project management, cloud services, and local systems via APIs. For example, it receives real-time onboarding, transfer, and offboarding events from the HR system and reflects them in the user information management and permission granting departments. It synchronizes authentication and permission information with external systems such as cloud storage services and groupware to achieve single sign-on. The API linkage unit automatically parses the API specifications and authentication protocols of each system, performing the necessary data conversion and authentication processing. Specifically, this API linkage unit adopts an asynchronous event-driven architecture using a message queue system. It installs a sequence transformation neural network for data conversion. The input data to this sequence transformation neural network is a data object described in the unique format of the sending source system. This sequence transformation neural network encodes the input data into an intermediate representation vector. The output data from this sequence transformation neural network is a data object following a standardized common data model. This API linkage unit publishes the converted data to a specific topic on the message broker. This process eliminates tight coupling between systems, improving fault tolerance. This process prevents message loss even during traffic spikes. This system maximizes the throughput of data conversion processing, ensuring the scalability of the entire system.

[0056] Specific examples of prompts required for the generative AI to read in implementing this invention include: "Automatically extracting the minimum set of permissions required by the development department manager from past permission granting history and business content, and generating a recommended permission list"; "Extracting access patterns and permission change patterns that differ from the norm from audit records over the past year, and listing high-risk users"; "Automatically resetting the permissions of corresponding users in all systems based on transfer events received from the HR system, and revoking unnecessary permissions"; "Detecting new folder events in cloud storage services and automatically granting access permissions only to employees in the corresponding departments"; "Dynamically controlling access permissions to specific systems based on employee qualification information"; and "Analyzing audit records, extracting permission setting errors and personalization tendencies, and proposing corrective measures and the necessity of education," etc. By inputting these prompts into the generative AI, the AI ​​can achieve a variety of functions such as optimizing the overall system's operation, strengthening security, improving operational efficiency, eliminating personalization, accumulating proprietary technology, and enhancing audit response capabilities. Specifically, this system uses prompt engineering techniques to improve the inference accuracy of large language models. The input data to this system combines system prompts, contextual data containing past log vectors, and text sequences of user queries. This system employs a thought chain prompting technique to force the model to perform intermediate inference steps. The output data from this system is structured execution plan data that defines the order and parameters of the tasks to be executed. The system uses an orchestration engine to parse the execution plan data. The system calls the interfaces of each module sequentially according to the defined order. This processing suppresses the illusion phenomenon inherent in large language models, achieving deterministic system control. This processing accurately translates natural language-based instructions into a set of machine language-level commands that can be executed by the computer. This system enables highly automated applications.

[0057] (Implementation steps) Step 1: Unified Registration and Synchronization of User Information When a new employee joins a business assignment, detailed attribute information such as name, employee number, department, position, employment status, contact information, work start date, scheduled work end date, responsible project, scheduled system usage, past transfer history, skill set, qualification information, and emergency contact information are uniformly registered in the database within the User Information Management Department. Furthermore, it integrates with external systems such as the HR system, contract management system, and project management system via API to receive real-time event information such as employee onboarding, transfers, departures, contract updates, and contract terminations, automatically synchronizing and updating user information. For example, upon receiving new assignment information from the project management system, the responsible project information is reflected immediately. Specifically, this system receives requests from external systems via a network hook using the Hypertext Transfer Protocol (HTTP). The input data to this system is a payload describing the detailed content of the event. This system uses regular expressions to validate the format of each data field within the payload. If a validation error is detected, the system returns an error code to the sending source. The system normalizes successfully validated data and generates an update query for the database. The output data from this system is an update command described using Structured Query Language (SCL). This system sends update commands to the database engine. It then sends an update completion event to the message broker to trigger subsequent processing. This process prevents the intrusion of illegal data at the edge, maintaining database consistency. It also eliminates the risk of business downtime caused by data inconsistencies between systems. Furthermore, the system automates data entry to prevent the waste of human resources.

[0058] Step 2: Automated Minimum Permission Strategy Developed via Generative AI. In the permission management department, generative AI is utilized to comprehensively analyze each user's department, position, business content, past permission granting performance, system usage history, project information, skill set, and qualification information to automatically develop a minimum permission granting strategy. For example, granting the development department manager administrative permissions to the source code management system and access to the development environment, while granting CS department employees only viewing permissions to the customer information management system, the AI ​​automatically determines the optimal permission set. Specific examples of prompts for the generative AI to read include: "Automatically extract the minimum permission set required by the development department manager from past permission granting history and business content, and generate a recommended permission list," and "Extract only the required permissions from the CS department employees' business content and past usage history, and formulate a minimum permission granting strategy." Specifically, this permission management department executes a deep neural network with an attention mechanism. The input data to this deep neural network is a tensor combining a 512-dimensional user attribute vector and a system requirement matrix. This deep neural network performs weighted calculations on the dependencies between the elements within the tensor. The output data from this deep neural network is a one-dimensional array storing probability scores representing the necessity of each permission. This permission grants the management department the ability to extract permissions with a probability score of 0.9 or higher. This permission also grants the management department the ability to dynamically generate access control list definition files based on the extracted permission sets. Furthermore, this permission allows the management department to deploy the generated definition files to various target systems. This process completely eliminates human error caused by manual settings. This process strictly and automatically applies the principle of least privilege. This system simplifies compliance verification in security audits and drastically reduces implementation costs.

[0059] Step 3: Automatically Assign, Modify, and Revoke Permissions via Event Triggers. Through API linkage with the HR and contract management systems, the system receives real-time event information such as employee onboarding, transfers, departures, contract updates, and contract terminations, and immediately assigns, modifies, or revokes the corresponding user's permissions. For example, in the case of a transfer from the Development Department to the CS Department, the AI ​​detects the transfer event, automatically revoking the permissions granted in the Development Department and only granting new permissions in the CS Department. Upon departure or contract termination, the AI ​​automatically deletes the corresponding user's account from all relevant systems, revoking all permissions. Specific examples of prompts for the generative AI to read include: "Based on the transfer event received from the HR system, automatically reset the corresponding user's permissions across all systems and revoke unnecessary permissions," and "Upon detecting a departure event, delete the account from the relevant systems and leave an audit record." Specifically, this system adopts an event-driven architecture. The system has a parallel computing cluster using image processing semiconductors. The API linkage unit receives JSON-formatted event payloads from external systems. The processing unit parses the event data. The processing unit tokenizes the text data. The processing unit removes unnecessary symbols. This generative AI processing unit is equipped with a large-scale language model with billions of parameters. This model has a structure with multiple stacked Transformer layers. This model uses a self-attention mechanism to compute dependencies between tags. This generative AI processing unit receives multi-dimensional word embedding tensors as input data. This input data includes vector representations of text such as "User A's affiliation changed from Development Department to CS Department." This input data includes vector representations of text such as "User B's contract has ended." This input data includes vector representations of text such as "User C has been assigned to a new project." This generative AI processing unit generates structured data as output data, indicating permission changes. This output data includes key-value pairs such as "System X's access permission: delete." This output data includes key-value pairs such as "System Y's access permission: append." This output data includes key-value pairs such as "System Z's account status: invalidate." This permission grants the management department the ability to parse the structured data output. This permission grants the management department the ability to issue permission change commands to the access control interfaces of each target system. This permission grants the management department the ability to send the execution results of the commands to the audit record management department. This model uses past permission change history for pre-learning. The learning process updates weights by minimizing the cross-entropy error function. The learning process uses the gradient descent algorithm. This architecture can reflect permissions instantly without manual administrator intervention. This architecture completely prevents the omission of permission revocations. This architecture significantly reduces the security risk of information leakage. This architecture improves the overall system's processing throughput. This technology is applicable as a unified ID management infrastructure in multinational corporations with tens of thousands of employees. This technology is suitable for access management systems in financial institutions that frequently undergo organizational changes.This system dynamically interprets complex business rules through vector operations in high-dimensional space. It offers greater flexibility than previous static, rule-based processing methods.

[0060] Step 4: Detailed Collection and Tamper-Proofing of Audit Logs In the Audit Log Management Department, all account operations (new registration, information updates, deletions), permission changes (granting, modifying, revoking), system access history (login, logout, operation content), data synchronization history with external systems via API, terminal information (IP address, terminal identifier, browser information, etc.), and time information are recorded in detail. For important permission changes or account deletion operations, multi-factor authentication by the administrator is mandatory, and electronic signatures are assigned to the operation records to prevent subsequent tampering or denial. Audit logs are backed up regularly and also have archiving capabilities for long-term storage. Specifically, this Audit Log Management Department has an advanced data processing infrastructure linked to distributed ledger technology. This system has multiple storage nodes connected via a high-speed internal bus. This Audit Log Management Department works in conjunction with the Generative AI Processing Department to parse logs. This processing department normalizes the raw log data. This processing department unifies the timestamps to Coordinated Universal Time (UTC). This generative AI Processing Department installs a natural language processing model. This generative AI Processing Department receives time-series arrays of audit logs as input data. This input data includes, for example, a numerical vector of login time. This input data includes, for example, a string vector of the target IP address accessed. This input data includes, for example, a text vector of the operation command. This generative AI processing unit generates a summary text of the logs and a risk score as output data. This output data includes, for example, a floating-point number with a risk score of 0.95. This output data includes, for example, a warning label indicating a high probability of unauthorized access. This output data includes, for example, a flag recommending the forced termination of the corresponding session. This audit log management unit compares the output risk score with a threshold. This audit log management unit sends an immediate alert to the administrator if the score exceeds the threshold. This audit log management unit applies a hash function to all log data. This audit log management unit records the generated hash value in the blockchain network. Through this processing, the integrity of the logs is mathematically proven. This model uses normal log patterns as learning data. This learning process employs a self-supervised learning method. This learning process optimizes the network weights by minimizing reconstruction errors. This structure instantly identifies minute anomalies that are difficult to detect by the human eye from massive amounts of log data. This structure brings a significant technical effect of reducing the workload of auditing operations. This structure achieves data management that strictly meets compliance requirements. This technology is applicable to electronic medical record systems in medical institutions requiring stringent auditing. It is also suitable for confidential information management systems in government agencies. This system goes beyond simple data storage, proactively performing semantic parsing of execution logs. Furthermore, it provides advanced security monitoring, distinct from traditional simple string retrieval.

[0061] Step 5: Anomaly Detection and Risk Alerts via Generative AI. In the generative AI processing unit, diverse data such as user information, permission granting history, audit logs, system utilization, and event information from external systems are comprehensively analyzed. Anomaly detection algorithms are used to detect unusual access and permission changes in real time. For example, if a user who doesn't normally access the system accesses an important system late at night, or if the same user repeatedly changes permissions in multiple systems within a short period, or if a former user's account is misused, the system automatically extracts abnormal patterns and notifies the administrator of the risks. Specific examples of alerts read by the generative AI include: "Extract unusual access and permission change patterns from the past year's audit logs and list high-risk users," and "Immediately notify the administrator of alerts in the event of unusual permission changes and access." Specifically, this generative AI processing unit adopts an autoencoder architecture based on Long Short-Term Memory (LSTM) networks. The system has a tensor operation unit dedicated to processing time-series data. This processing unit maps diverse log data to the same dimensional space. This processing unit performs one-hot encoding on categorical data. This processing unit standardizes continuous values ​​to have zero mean and unit variance. The generative AI processing unit receives time-series access log tensors as input data. This input data includes, for example, binary flags representing access at 2 AM. This input data includes, for example, numerical data on the number of login attempts from IP addresses different from the usual ones. This input data includes, for example, counts of permission changes within a short period. The generative AI processing unit generates an anomaly score based on reconstruction error as output data. This output data includes, for example, a floating-point number such as an anomaly score of 0.88. This output data includes, for example, a vector representing the contribution of IP addresses as anomalies. This output data includes, for example, account lockout indicators as recommended responses. This processing unit determines whether the output anomaly score exceeds a pre-set dynamic threshold. If it detects an exceedance, this processing unit immediately invalidates the session token of the corresponding account. This processing unit also displays a warning on the security operations team's dashboard. This model learns only using past normal access history. This learning process uses a backpropagation algorithm along time. This learning process uses mean squared error as the loss function. This architecture blocks unknown network attacks and internal violations in real time. This architecture provides a preventative technical advantage, preventing damage from escalating before it occurs. It reduces false alarm rates and alleviates the fatigue of those responsible for its use. This technology is applicable to countermeasures against unauthorized logins on large-scale e-commerce platforms. It is also applicable to infrastructure monitoring systems for cloud service providers. This system performs dynamic, behavior-based detection, rather than static, signature-based detection. This system achieves objective and highly accurate risk assessments that do not rely on human intuition.

[0062] Step 6: System Synchronization and Single Sign-On via API Linkage. The API Linkage Department integrates data with external systems such as HR, contract management, project management, cloud services, and local systems via APIs. It synchronizes authentication and permission information with external systems like cloud storage services and groupware to achieve single sign-on. AI automatically parses the API specifications and authentication protocols of each system, performing the necessary data conversion and authentication processing. For example, if a new folder is created in a cloud storage service, the AI ​​detects this event and automatically grants access permissions only to employees in the corresponding department. Specific examples of prompts read by the generative AI include: "Detecting a new folder creation event in the cloud storage service and automatically granting access permissions only to employees in the corresponding department," and "Real-time monitoring of the authentication status and usage of each system; immediately locking the corresponding account and notifying the administrator of an alert if abnormal authentication requests or illegal token usage are detected." Specifically, this API Linkage Department uses a graph neural network dependency parsing engine. This system uses an asynchronous message queue protocol to control inter-system communication. This processing department converts data obtained from various systems into a common data model. This processing department supplements missing data values. This generative AI processing unit receives inter-system dependency graph data as input data. This input data includes, for example, edge weights representing the frequency of interaction between system A and system B. This input data includes, for example, node features representing the user's department. This input data includes, for example, time-series numerical data representing past authentication success rates. This generative AI processing unit generates the optimal synchronization timing and the required permission set as output data. This output data includes, for example, a timestamp representing the recommended time for synchronization execution. This output data includes, for example, a list of the minimum permissions to be granted. This output data includes, for example, a deletion flag for unnecessary permissions. This API linkage unit automatically generates API requests to each external system based on the output permission set. This API linkage unit sends the generated requests in an appropriate order. This API linkage unit monitors the response codes from each system. This model learns the inter-system communication logs as a graph structure. This learning process simultaneously optimizes node classification and link prediction tasks. This learning process updates the weights of the graph convolutional network. This structure automatically eliminates complex and intertwined permission inconsistencies between multiple systems. This structure significantly reduces the technical barriers to implementing single sign-on. This structure optimizes network traffic and minimizes communication latency. This technology is applicable to the IT environments of modern enterprises utilizing most SaaS applications. It is also suitable for system integration projects accompanying enterprise mergers or acquisitions. This system performs not simple data replication, but semantic data transformation and synchronization. It autonomously controls inter-system interactions on a scale that cannot be manually set or managed by humans.

[0063] Step 7: Application, Approval, and Information Update via User Interface. The user interface provides a screen for employees and administrators to confirm their account information and permissions, and to submit and receive approvals. When an employee applies to use a new business system, they send the application information from the terminal to the server. The server-side generative AI analyzes the application content, automatically determines whether the required permissions are appropriate, and returns approval or rejection feedback to the terminal. If an employee updates their skills or qualifications, they input this information from the terminal, and the AI ​​automatically grants access to the corresponding system. Administrators can view audit logs and risk alerts issued by the AI ​​from the terminal and instruct additional investigations and corrective actions as needed. Specifically, this user interface functions as a conversational agent assembled with a natural language processing model. This system has an edge caching mechanism to optimize front-end rendering. This processing unit normalizes the natural language text input by the user. This processing unit removes stop words from the text. This generative AI processing unit installs a bidirectional encoder representation model. This generative AI processing unit receives word embedding vectors of the user's application text as input data. This input data includes, for example, permission request text for participating in a new project. This input data includes, for example, a string representing the system name to be appended. This input data also includes, for example, numerical data indicating the period of use. This generative AI processing unit generates an appropriateness score and an automatic approval flag for the application as output data. This output data includes, for example, a floating-point number representing an appropriateness score of 0.92. This output data includes, for example, a Boolean value indicating whether automatic approval is possible. This output data includes, for example, text indicating deficiencies. This user interface unit dynamically changes the displayed content based on the output appropriateness score. This user interface unit delegates processing to the permission granting management unit when the automatic approval flag is true. This user interface unit prompts the user for additional input if there is insufficient information. This model is fine-tuned using historical application and approval data. This learning process adjusts attention weights to minimize classification error. This learning process uses the Adam optimizer. This structure significantly reduces the time required for administrator approval tasks. This structure results in minimizing user waiting time and improving business efficiency. This structure eliminates ambiguity in application content, achieving accurate permission granting. This technology is applicable to large-scale development centers where thousands of permission applications occur daily. This technology is applicable to asynchronous business process management in remote work environments. The system accurately captures not only standardized form input but also freely expressed intent. It reproduces, on a computer, the context-aware, flexible judgments made by a human administrator.

[0064] Step 8: Enhance Security Through Authentication Processing. In the authentication processing unit, user and administrator authentication is implemented when accessing the system. Multi-factor authentication, including combined password authentication, one-time password authentication, biometric authentication, and terminal authentication, is used to prevent identity impersonation and unauthorized access. When an administrator makes important permission changes, one-time password authentication or fingerprint authentication is required on the terminal, and the authentication result is sent to the server to authorize the operation. Authentication history is sent to the audit record management department to record all authentication events. Specifically, this authentication processing unit installs a biometric information parsing engine using a convolutional neural network. This system has a lightweight neural network for inference on the edge device side. This processing unit preprocesses the biometric information obtained from cameras or sensors. This processing unit removes noise from the image. This processing unit adjusts the contrast. This authentication processing unit receives image tensors of biometric information as input data. This input data includes, for example, an array of RGB pixel values ​​for a facial image. This input data includes, for example, a binarized image of a fingerprint ridge pattern. This input data includes, for example, infrared image data of a vein pattern. This authentication processing unit generates a similarity score with the registration data as output data. This output data includes, for example, a floating-point number such as a cosine similarity of 0.99. This output data includes, for example, a Boolean value indicating successful authentication. This output data includes, for example, a floating-point number indicating the probability of identity impersonation. This authentication processing unit issues a session token only if the output similarity score exceeds a strict threshold. This authentication processing unit requests additional authentication elements if the probability of identity impersonation is determined to be high. This authentication processing unit temporarily locks the account in the event of consecutive authentication failures. This model uses a biometric dataset of millions of people for pre-learning. This learning process uses a triplet loss function to bring the feature vectors of the same person closer together. This learning process updates the weights in a way that moves the feature vectors of different people further apart. This structure physically makes unauthorized access due to password leakage or theft impossible. This structure achieves the technical effect of maintaining a very high level of security without compromising user convenience. This structure reduces the authentication processing latency to milliseconds. This technology is suitable for access control in research and development departments that handle highly confidential information. This technology is suitable for privileged access management of the core systems of financial institutions. This system does not perform simple pattern matching, but rather performs identity verification through distance calculation in a high-dimensional feature space. This system provides robust certification against changes over time and variations in ambient light.

[0065] (specific use case) For example, in telecommunications or IT-related companies, when an employee is newly assigned to a large-scale project, the User Information Management Department uniformly registers detailed attribute information such as the employee's name, employee number, department, position, employment status, contact information, work start date, scheduled work end date, responsible projects, scheduled system usage, past transfer history, skill set, qualification information, and emergency contact information into a database. This is achieved through API integration with external systems such as HR, contract management, and project management systems, receiving real-time information on employee onboarding, transfers, departures, contract updates, and contract terminations, and automatically synchronizing and updating user information. For instance, upon receiving new assignment information from the project management system, the department immediately reflects the responsible project information. The Permissions Management Department uses generative AI to comprehensively analyze the employee's department, position, work content, past permission granting performance, system usage history, project information, skill set, and qualification information to automatically formulate a least privilege granting strategy. For example, granting the development department manager administrative privileges to the source code management system and access to the development environment, as well as full access to project management tools, while granting CS department employees only viewing privileges to the customer information management system and input privileges to the consultation response system, AI automatically determines the optimal set of permissions. Furthermore, AI continuously learns from past permission granting history and access logs, automatically extracting exception patterns where even employees in the same position require additional permissions when assigned to specific projects, flexibly updating permission granting rules. Through API integration with the HR system, it receives real-time information on employee onboarding, transfers, departures, contract updates, and contract terminations, instantly allocating, changing, or revoking permissions for corresponding users. For example, in the case of a transfer from the development department to the CS department, AI detects the transfer event, automatically revoking permissions granted in the development department and only granting new permissions in the CS department. Upon departure or contract termination, AI automatically deletes the corresponding user's account from all relevant systems, revoking all permissions. The audit record management department records all account operations, permission changes, system access history, data synchronization history with external systems via API integration, terminal information, and time information in detail. For significant permission changes or account deletions, multi-factor authentication by the administrator is mandatory. Electronic signatures are assigned to these records to prevent subsequent tampering or denial. Audit logs are backed up regularly and also have archiving capabilities for long-term storage. The generative AI processing unit comprehensively analyzes diverse data, including user information, permission granting history, audit logs, system utilization, and event information from external systems. Using anomaly detection algorithms, it detects unusual access and permission changes in real time. For example, it automatically extracts abnormal patterns and notifies the administrator of risks in cases such as users who don't normally access important systems late at night, the same user repeatedly changing permissions in multiple systems within a short period, or the unauthorized use of a former user's account.The API Integration Department integrates data with external systems such as HR, contract management, project management, cloud services, and local systems via APIs. It synchronizes authentication and permission information with external systems like cloud storage services and groupware to achieve single sign-on. AI automatically parses the API specifications and authentication protocols of each system, performing necessary data conversion and authentication processing. For example, if a new folder is created in a cloud storage service, AI detects the event and automatically grants access permissions only to employees in the relevant department. The User Interface Department provides screens for employees and administrators to confirm their account information and permissions, submit applications, and receive approvals. When an employee applies to use a new business system, the application information is sent from the terminal to the server. The server-side generative AI parses the application content, automatically determines the appropriateness of the required permissions, and returns approval or rejection feedback to the terminal. The Authentication Processing Department implements personal authentication for users and administrators accessing the system, combining multi-factor authentication such as password authentication, one-time password authentication, biometric authentication, and terminal authentication to prevent identity impersonation and unauthorized access. When administrators make important permission changes, one-time password authentication or fingerprint authentication is required on the terminal, and the authentication result is sent to the server to authorize the operation. Authentication history records are sent to the audit record management department to record all authentication events. Specifically, the entire system is uniformly controlled by a multi-agent reinforcement learning architecture. The system features a distributed database and a real-time stream processing engine. The processing unit processes data continuously flowing in from various subsystems in memory. The processing unit shards data to distribute the load. The generative AI processing unit receives environmental state vectors as input data. This input data includes, for example, numerical data representing the current load status of the entire system. This input data includes, for example, count data of the total number of user accesses. This input data includes, for example, integer values ​​of the number of security alerts occurring. The generative AI processing unit generates optimal resource allocation and access control policies as output data. This output data includes, for example, classification values ​​representing the access restriction level for a specific system. This output data includes, for example, flags indicating an increased monitoring level. This output data includes, for example, indications for extending the retention period of audit logs. The system applies the output policies to each subsystem in real time. The system dynamically changes the allocation of network bandwidth. The system automatically blocks specific communication ports when an anomaly occurs. The model learns through millions of rounds in a simulated environment. This learning process updates the action value function of each agent in a way that maximizes the reward function. This learning process employs Markov decision process-based optimization. This architecture achieves overall optimization impossible in individual system management. This architecture ensures stable system operation even during large-scale organizational changes or cyberattacks. This architecture maximizes security while drastically reducing operational costs. This technology is applicable to the core systems of global telecommunications operators with tens of thousands of servers and hundreds of thousands of users.This technology is applicable to the unified management of large-scale data centers with mission-critical requirements. The system autonomously controls complex inter-system interactions that exceed the limits of human cognition. This system provides an innovative approach that fundamentally overturns previous siloed system management methods.

[0066] Specific examples of prompts read by generative AI include: "Automatically extracting the minimum set of permissions required by the development manager from past permission granting history and business content, and generating a recommended permission list"; "Extracting only the necessary permissions from the business content and past usage history of CS department employees, and formulating a minimum permission granting strategy"; "Automatically resetting the permissions of the corresponding users in all systems based on transfer events received from the HR system, and revoking unnecessary permissions"; "Deleting the account from all related systems and leaving an audit record when a resignation event is detected"; "Extracting access patterns and permission change patterns that are different from the usual from the audit records of the past year, and listing high-risk users"; "Detecting new folder events in cloud storage services and automatically granting access permissions only to employees in the corresponding departments"; "Monitoring the authentication status and usage of each system in real time, and immediately locking the corresponding account and notifying the administrator when abnormal authentication requests or illegal token use are detected"; "Analyzing audit records, extracting permission setting errors and personalization tendencies, and proposing corrective measures and the necessity of education," etc. Specifically, this generative AI processing department installs a meta-learning model for automatically optimizing prompts. This system has a large capacity of memory for processing large-scale context windows. This processing unit performs syntactic parsing on the initial input prompts. It extracts the task's purpose and constraints. The meta-learning model receives a text vector describing the task as input data. This input data includes, for example, embedding vectors of instructions for permission retrieval tasks. This input data includes, for example, historical data of past successful prompts. This input data includes, for example, vector representations of the target system's specification text. The meta-learning model generates optimized prompt strings as output data. This output data includes, for example, detailed instructions enhanced with context. This output data includes, for example, strict specifications of the output format. This output data includes, for example, thought chain hints used to explicitly indicate the inference process. This generative AI processing unit sends an inference request to the main language model using the generated optimized prompts. This generative AI processing unit verifies the output from the main model. If the output does not meet the requirements, this generative AI processing unit regenerates the prompts and retrys. The meta-learning model uses a few-shot learning technique to learn the characteristics of the task from a few samples. This learning process uses a reward model to evaluate the effectiveness of the prompts for reinforcement learning. This learning process uses a policy gradient method to update the parameters of the prompt generator. This structure completely eliminates the labor of manually adjusting prompts by humans. This architecture delivers a significant improvement in the accuracy and stability of AI output. It ensures optimal inference results even for complex business requirements. This technology can be applied to automated infrastructures in diverse enterprise environments with a mix of business systems. It enables users without specialized knowledge to leverage advanced AI capabilities through interfaces.This system is not simply a string template, but rather it understands the semantics of the task and dynamically constructs commands. This system enables an advanced self-optimizing loop where AI controls itself.

[0067] (Application Example 1) The specific processing flow corresponding to Use Case 1 will be described below. The various parts of the system described below are implemented by the data processing device 12 and the intelligent device 14. Furthermore, the data processing device 12 is referred to as the "server" and the intelligent device 14 is referred to as the "terminal".

[0068] (Implementation methods for carrying out the present invention) As an embodiment of this invention, a system architecture is adopted that integrates and dynamically automates the management of physical and logical access permissions in a data center. This system includes a user information management department, a permission granting management department, an audit log management department, a generative AI processing department, an API linkage department, a user interface department, and an authentication processing department, each of which operates on a server or terminal. Specifically, this system adopts a hybrid architecture integrating edge computing and cloud computing. This system is connected to a group of IoT sensors deployed throughout the data center via a high-speed network. This processing department receives streaming data continuously transmitted from each sensor. This processing department performs noise filtering on the data. This generative AI processing department receives the temporal arrangement of sensor data as input data. This input data includes, for example, numerical data from temperature sensors in a server room. This input data includes, for example, dynamic detection flags from surveillance cameras. This input data includes, for example, timestamps of opening and closing history records of door access control systems. This generative AI processing department generates a comprehensive security status score for the data center as output data. This output data includes, for example, a floating-point number representing a comprehensive security score of 0.85. This output data includes, for example, an integer value representing a warning level. This output data includes, for example, 3D coordinate data of the area where the anomaly occurred. This system issues control commands to each device based on the output score. It controls the air conditioning in the corresponding area upon anomaly detection. It flashes lights and sends a notification to security personnel upon suspicious personnel detection. The model is pre-trained using an autoencoder with normal environmental data. The learning process optimizes the network weights to minimize reconstruction errors from the input data. This learning process performs representation learning of the latent space. This architecture eliminates the boundaries between physical and logical spaces, enabling comprehensive security monitoring. This architecture dramatically improves response speed in the event of an incident. This architecture maximizes data center utilization efficiency and minimizes downtime. This technology can be applied to mainframe application centers of financial institutions requiring extremely high availability. This technology can be applied to the physical security management of government data centers handling state secrets. This system is not simply a log aggregation but interprets dynamic changes in the physical environment in real time. This system provides the flexibility for independent expansion of each functional module through a microservice architecture.

[0069] The User Information Management Department centrally manages all users entering and exiting the data center in the database. This includes a wide range of attributes such as name, ID, role, company affiliation, employment status (currently employed, former, visitor, contract terminated, etc.), contact information, shift schedule, area of ​​responsibility, past entry / exit history, qualification information, emergency contact information, health status, work schedule, responsible equipment, access history, and room access control usage history. For example, if an external contractor only needs to enter the server room during a specific period, detailed registration information can be provided, including the period, area of ​​responsibility, work content, scheduled work time, person in charge, required qualifications, availability of work procedures, and emergency contact information. Similarly, if an equipment inspector temporarily enters the data center for an annual inspection, the system can record the scheduled inspection date, area of ​​responsibility, required equipment list, inspection content, and deadline for submitting the inspection report. For example, when cleaning staff enter a specific area at fixed times each week, this can be registered as a regular schedule in the work schedule and managed in conjunction with their entry and exit history. Similarly, when temporary workers enter for emergency response, the system can record the reason for the emergency, the content of the response, the start and end times of the response, the person in charge, and the reporting obligations afterward. Specifically, this user information management department uses a graph database to model complex relationships. The system maintains a relational graph consisting of nodes and edges in memory. This processing department maps diverse user attribute information to a multidimensional vector space. This processing department quantifies textual information using word embedding technology. This generative AI processing department uses an autoencoder-based feature extraction model. This generative AI processing department receives multidimensional vectors of user attributes as input data. This input data includes, for example, continuous values ​​such as age and years of continuous work experience. This input data includes, for example, categorical data such as qualifications held. This input data includes, for example, counts of past room entries and exits. This generative AI processing department generates a latent representation vector of the user as output data. This output data includes, for example, a 64-dimensional latent feature vector. This output data includes, for example, group IDs used for clustering. This output data also includes, for example, a list of similar users. The user information management department uses the output latent vectors to calculate the initial allocation of access permissions. The user information management department refers to the permission settings of similar users to suggest recommended settings. The user information management department recalculates the latent vectors when changes in attribute information are detected. This model is trained using data from tens of thousands of users through unsupervised learning. This learning process minimizes the error between the input and output vectors. This learning process optimizes the data distribution in the latent space. This structure manages a diverse user group with various employment types and roles using a unified standard. This structure provides the technical effect of completely eliminating errors from manual attribute input and setting omissions.This architecture dramatically improves database retrieval speed. This technology can be applied to access control systems in large-scale manufacturing plants with thousands of employees. It can also be applied to personnel management systems in construction sites with multiple external suppliers. This system transcends the framework of simple relational databases, parsing the semantic relationships between data. This system achieves intuitive grouping and access control through similarity calculations in high-dimensional space.

[0070] The access control department leverages generative AI to comprehensively analyze each user's role, affiliated company, business content, work plan, qualification information, past access performance, scheduling information, responsible area, health status, work supervisor, work content, and emergency response history. It automatically assigns the minimum and optimal logical access permissions (e.g., server management, network settings, monitoring system operation, equipment management system operation, log viewing, setting changes, and fault response permissions) and physical access permissions (e.g., server room, power room, monitoring room, storage area, air conditioning equipment room, electrical equipment room, network equipment room, common areas, and rest rooms). For example, maintenance technicians are granted access to the server room and monitoring room, as well as server management system operation permissions; external contractors are granted access to the server room only during work hours and temporary setting change permissions; security personnel are granted access to all areas and full access to the monitoring system; cleaning personnel are granted physical access limited to the cleaning area; equipment inspectors are granted physical access to their designated inspection area and access to the equipment management system; and audit personnel are granted access to all areas and log viewing only during audit periods. For example, the AI ​​learns from past access history and anomaly detection history to instantly detect unusual actions such as "external contractors attempting to enter power rooms that are not normally accessed," "maintenance technicians attempting to enter server rooms late at night," "cleaning personnel attempting to enter areas outside their assigned areas," "equipment inspectors attempting to enter areas outside their scheduled inspection areas," and "temporary workers remaining in the area for extended periods after an emergency response has ended," and alerts the administrator accordingly. For example, based on qualification information, access to specific areas or systems can be granted only to users with specific advanced security qualifications. For example, physical access can be temporarily suspended and the administrator notified if a user is determined to be in poor health. For example, access can be temporarily restricted for workers when the person in charge of a task is absent. Specifically, this access control system uses an ensemble learning model that combines decision trees and neural networks. The system is connected to the physical access control device via a low-latency network. The processing unit combines the user's static attributes with dynamic environmental information. The processing unit normalizes the combined data and generates feature vectors. The generative AI processing unit receives the combined vector of user features and environmental features as input data. This input data includes, for example, the user's latent representation vector. This input data includes, for example, the current time. This input data includes, for example, the importance score of the target area. This generative AI processing unit generates an access permission probability as output data. This output data includes, for example, a floating-point number with a permission probability of 0.95. This output data includes, for example, a list of permissions that should be granted. This output data includes, for example, numerical data representing the accessibility period.This permission grants the management department the authority to send an unlock command to the access control system when the probability of an output permission exceeds a threshold. This permission also grants the management department the authority to initiate additional approval processes when the probability of permission is low. Furthermore, this permission grants the management department the authority to immediately suspend full access upon receiving a flag indicating a deterioration in health status. This model uses past access approval history and event logs for supervised learning. This learning process updates the weights of each model to minimize cross-entropy error. This learning process uses a gradient boosting algorithm. This structure ensures complete synchronization of physical and logical access. This structure delivers the technical effect of strictly applying the principle of least privilege and minimizing insider threats. This structure enables dynamic control that dynamically revoks permissions in real time based on changes in circumstances. This technology can be applied to room access management in research institutes with high biosecurity levels. This technology can be applied to strict access control in critical facilities of nuclear power plants. This system does not rely on static access control lists but understands the context and makes judgments. This system processes thousands of access requests simultaneously with a speed and precision impossible for human security guards.

[0071] The Audit Records Management Department meticulously records all user account operations (new registration, information updates, deletion), permission changes (granting, modifying, revoking), physical room entry and exit records (room access control access time, access control ID, reason for entry / exit, accompanying person information, video recording of entry / exit, etc.), logical system access records (login, logout, operation content, operation terminal information, operation time, target system, details of operation content, operation result, error occurrence history, etc.), data synchronization history with external systems via API linkage, and terminal information (room access control ID, terminal IP address, terminal identifier, browser information, terminal security status, etc.) along with time information. For example, when maintenance technicians enter the server room, log into the server management system, make settings changes, and implement fault handling, this series of physical and logical access and operation history records are recorded chronologically. Similarly, if an external contractor attempts to enter a room outside of work hours, the attempt record and the access control authentication failure history are also meticulously recorded. For example, when an equipment inspector submits an inspection report after an inspection, this submission history is also saved as an audit record. For important permission changes or account deletion operations, multi-factor authentication by the administrator is required, and the operation record is electronically signed to prevent subsequent tampering or denial. Audit records are backed up regularly and also have archiving capabilities for long-term storage. For example, all permission change history, room entry / exit history, operation history, and anomaly detection history from the past 5 years are securely stored for rapid retrieval and extraction as needed. Specifically, this audit record management department adopts an architecture integrating a time-series database and a natural language generation model. This system has a distributed storage cluster for processing petabytes of data. This processing unit parses log data in various formats. This processing unit converts the data into structured JSON format. This generative AI processing unit receives tensors of structured log data as input data. This input data includes, for example, timestamps of the operation date. This input data includes, for example, the operator's ID number. This input data includes, for example, string data of the executed command. This generative AI processing unit generates human-readable audit report text as output data. This output data includes text such as daily security summaries. This output data includes highlighted text such as anomalous operations. This output data includes text such as recommended improvement measures. The Audit Record Management Department digitizes the output reports and automatically sends them to the administrator. The Audit Record Management Department performs Merkle tree hash calculations on all log data. The Audit Record Management Department periodically writes the root hash to an external public blockchain. This model is fine-tuned using pairs of past audit reports and raw log data. This learning process maximizes the BLEU score between the generated text and the correct answer text.This learning process optimizes the weights of the Transformer's decoder layer. This architecture transforms massive amounts of tedious log data into meaningful intelligence. This architecture dramatically reduces the time required for audit responses from weeks to minutes. This architecture mathematically makes data tampering impossible and guarantees evidentiary strength. This technology can be applied to the accounting systems of banks undergoing rigorous audits by the Financial Services Agency. This technology can be applied to trajectory management systems designed to comply with the internal control reporting systems of publicly traded companies. This system is not merely a data storage device, but functions as an intelligent auditor that autonomously generates reports. This system minimizes long-term storage costs through data compression algorithms and hierarchical storage management.

[0072] The generative AI processing unit comprehensively analyzes diverse data, including user information, permission granting history, audit logs, system utilization, event information from external systems, health status, work plans, emergency response history, equipment inspection history, and cleaning history. Using anomaly detection algorithms, it detects unusual access and permission changes in real time. For example, it automatically identifies abnormal patterns such as users who don't normally access important areas late at night, external contractors attempting to enter power rooms, unauthorized use of accounts belonging to former employees, cleaning staff entering areas outside their assigned areas, equipment inspectors entering areas outside their scheduled inspection areas, and temporary workers remaining in the area for extended periods after an emergency response has ended. It then notifies administrators of these risks. For instance, it analyzes administrator operation history and past audit logs to identify errors in permission settings and tendencies towards personalization, automatically reporting risks such as "frequent permission granting errors caused by specific administrators," "frequent abnormal access by specific external contractors," and "concentrated abnormal access in specific areas." It can also propose corrective measures and emphasize the necessity of training. Furthermore, AI analysis of user movement and permission transfer history across multiple data center sites can automatically evaluate and adjust the consistency of security policies across these sites. For example, AI analysis of equipment inspection and cleaning history records can detect omissions and illegal operations, and report to administrators. Specifically, this generative AI processing unit is equipped with a multimodal Transformer architecture. The system has a dedicated GPU cluster for simultaneously processing image and text data. This processing unit extracts image streams from surveillance cameras frame by frame. This processing unit converts text logs into word embedding vectors. This generative AI processing unit receives a combined tensor of image data and text logs as input data. This input data includes, for example, an arrangement of frame images from surveillance cameras. This input data includes, for example, authentication log text from room access control systems. This input data includes, for example, time-series numerical data from environmental sensors. This generative AI processing unit generates a comprehensive risk assessment result as output data. This output data includes, for example, classification values ​​indicating high or low risk levels. This output data includes, for example, the start and end times of abnormal actions. This output data includes, for example, bounding box coordinates of suspicious individuals. This generative AI processing unit immediately activates an alarm system when the output risk level reaches the highest value. This generative AI processing unit forcibly locks the electronic locks in the corresponding area. This generative AI processing unit backs up relevant video data to prohibited coverage areas. The model uses self-supervised learning pairs of images and logs of normal business landscapes. The learning process uses contrastive learning to approximate the latent representations of images and text. This learning process optimizes the decision boundary for anomaly detection. This architecture directly detects physically anomalous actions from images that cannot be discerned from text logs alone. This architecture delivers a dramatic reduction in false positives by multiplying multiple data sources.This structure automatically levels the deviations in security levels between network points. This technology can be applied to unified security operations centers of global enterprises spanning multiple countries. This technology can be applied to remote monitoring systems in unmanned smart factories. This system completely replaces the previous practice of human monitors watching multiple displays. This system performs mathematically intuitive risk assessments through multimodal computation in high-dimensional space.

[0073] The API Linkage Department integrates data with external systems such as HR, Access Control, Equipment Management, Cloud Services, Local Systems, Health Management, and Emergency Notification Systems via APIs. For example, it receives information on onboarding, transfers, and departures from the HR system; real-time room entry / exit information from the Access Control system; inspection schedules from the Equipment Management system; health status information from the Health Management system; and emergency response information from the Emergency Notification system, instantly reflecting this information to the User Information Management Department and the Access Control Department. For instance, it synchronizes authentication and permission information with external systems such as cloud storage services and surveillance camera systems to achieve single sign-on and unified access control. For example, upon receiving inspection schedule information from the Equipment Management system, it automatically adjusts the access permissions of corresponding inspectors. For example, upon receiving contract termination information from an external contractor, it immediately revokes all physical and logical access permissions and saves the entire operation history as an audit log. For example, upon receiving abnormal health information from the Health Management system, it temporarily suspends the access permissions of the corresponding user and notifies the administrator. Specifically, this API Linkage Department installs a routing optimization model based on reinforcement learning. This system also features an event-driven message queue based on a publish-subscribe model. This processing unit buffers asynchronous messages from various external systems. This processing unit determines message priority. This generative AI processing unit receives network traffic state vectors as input data. This input data includes, for example, the response time of each API endpoint. This input data includes, for example, an integer value indicating the number of messages in the queue. This input data includes, for example, numerical data indicating the error rate. This generative AI processing unit generates optimal routing paths and retry intervals as output data. This output data includes, for example, the IP address of the server to be prioritized. This output data includes, for example, a backoff time in milliseconds. This output data includes, for example, a timeout threshold. This API linkage unit dynamically changes the load balancer settings based on the output routing paths. This API linkage unit reallocates traffic to prevent system congestion. This API linkage unit activates a circuit breaker function to automatically disconnect endpoints with persistent errors. This model uses traffic patterns in a simulated environment for reinforcement learning. This learning process sets a reward function to maximize the throughput of the entire system. This learning process uses the Q-learning algorithm to update action values. This structure prevents external system failures and delays from affecting this system. This structure brings the technical effect of maximizing the real-time performance of data synchronization. This architecture eliminates single points of failure in complex inter-system interactions. This technology can be applied to large-scale cloud-native applications with thousands of interconnected microservices. This technology can be applied to the data interconnection infrastructure of real-time critical financial trading systems. This system does not rely on static routing tables but adapts to dynamic network state changes. This system autonomously performs complex flow control that cannot be designed by humans.

[0074] The user interface provides screens for users and administrators to confirm their account information and permissions, and to submit and receive applications. For example, when an external contractor requests an extension of the work period, the application information is sent from the terminal. The generative AI on the server side analyzes the application content, automatically determines whether the required permissions are appropriate, and returns feedback to the terminal regarding approval or rejection. For example, when an equipment inspector obtains new qualifications, the information is entered from the terminal, and the AI ​​automatically grants access permissions to the corresponding equipment. Administrators can view audit logs and risk alerts issued by the AI ​​from the terminal and instruct additional investigations and corrective actions as needed. For example, when temporary workers apply to leave the site after an emergency response, the AI ​​automatically evaluates the response content and duration of stay, and alerts the administrator if anything is suspicious. Specifically, this user interface employs an advanced front-end architecture that integrates gaze estimation and operation prediction models. This system uses the WebSocket real-time bidirectional communication protocol. This processing unit collects mouse movement and keystroke events from the user's browser. This processing unit converts the collected event data into time-series tensors. This generative AI processing unit receives the user's operation history and mouse trajectory data as input data. This input data includes, for example, the temporal arrangement of the XY coordinates of the mouse pointer. This input data includes, for example, the ID of the clicked button. This input data includes, for example, numerical data such as the screen dwell time. This generative AI processing unit generates a prediction result of the information needed next as output data. This output data includes, for example, the URL of data to be pre-fetched. This output data includes, for example, text of recommended actions. This output data includes, for example, suggestion strings for input assistance. This user interface unit prefetches data in the background based on the output prediction results. This user interface unit dynamically updates the screen rendering to improve motion sensing speed. This user interface unit automatically displays a help pop-up window when it determines that the user is hesitating. This model uses the operation logs of most users as a sequence transformation model for learning. This learning process minimizes the cross-entropy error when predicting the next action. This learning process uses an attention mechanism to assign weights to important operation history. This structure anticipates the user's thinking and instantly prompts the required information. This structure brings a dramatic reduction in the cognitive load required for application and approval processes. This structure provides an intuitive operating environment that prevents input errors and application omissions. This technology can be applied to a company-wide portal used by employees with varying IT literacy levels. This technology can be applied to dashboards in security operations centers that handle high-speed, time-sensitive incident responses. The system is not simply a static HTML display; it dynamically interprets user intent. This system pushes the boundaries of human-computer interaction.

[0075] The authentication processing unit implements personal authentication for users and administrators accessing the system, combining multi-factor authentication such as password authentication, one-time password authentication, biometric authentication, terminal authentication, facial authentication, and voice authentication to prevent impersonation and unauthorized access. For example, when an administrator makes important permission changes, one-time password authentication, fingerprint authentication, and facial authentication are required on the terminal, and the authentication result is sent to the server to authorize the operation. Authentication history is sent to the audit record management department, and all authentication events are recorded. For example, facial authentication and IC card authentication at room access control can also be combined to strengthen personal verification. Specifically, this authentication processing unit installs a fusion model integrating biometric authentication and behavioral authentication. This system has a quantized neural network for high-speed inference at the edge device side. This processing unit simultaneously acquires facial images from a camera and typing data from a keyboard. This processing unit extracts landmark feature points from the facial images. This generative AI processing unit receives facial images and dynamic keystroke data as input data. This input data includes, for example, the arrangement of facial landmark coordinates. This input data includes, for example, pressure data when a key is pressed. This input data includes, for example, the millisecond value of the typing interval. This generative AI processing unit generates a comprehensive identity verification score as output data. This output data includes, for example, a floating-point number indicating the identity of the user (0.99). This output data includes, for example, a floating-point number indicating the impersonation probability (0.01). This output data includes, for example, a Boolean value indicating a re-authentication request. This authentication processing unit only grants access if the output identity probability exceeds a threshold. This authentication processing unit provides additional voice authentication prompts if the score is in the boundary region. This authentication processing unit encrypts the authentication result and the extracted feature vector and sends them to the audit record management department. This model is pre-trained using a large-scale multimodal dataset. This learning process maps feature representations between different modalities to the same latent space. This learning process uses contrastive learning to maximize the margin between the identity and others. This structure can detect impersonation through behavioral features even if a single biometric information is forged. This structure achieves a technical effect that balances extremely high security levels with user convenience. This structure reduces the authentication time to milliseconds, preventing congestion at access control points. This technology can be applied to access control systems in government agencies handling state secrets. This technology can be applied to access control in the trading rooms of financial institutions where large sums of money move around. Instead of relying on static passwords, this system continuously verifies the user's physical and behavioral characteristics. This system provides an advanced authentication infrastructure that forms the foundation of a zero-trust security model.

[0076] In this way, in the embodiments of the present invention, the physical access management and logical system access management in the data center are integrated, which can realize a highly advanced unified access management system that can flexibly and securely meet a variety of application needs on site. This is achieved through dynamic and optimal permission management by generative AI, real-time detection of abnormal behavior, full operation auditing, automatic synchronization through API linkage with external systems, permission control that takes into account health status and emergency response history, access control that utilizes qualification information and job manager information, single sign-on, security enhancement through multi-factor authentication, and dynamic permission granting that takes into account job plans and inspection history. For example, applications that grant external contractors server room access only during operation periods and automatically revoke it after the period ends; applications that extract access history of users entering the server room late at night and detect abnormal patterns; applications that immediately invalidate the logical and physical access rights of departing users and save all operation history as audit logs; applications that detect equipment inspectors entering areas outside their designated inspection areas and immediately notify the administrator; applications that automatically suspend access rights when health is abnormal and re-grant them upon recovery; applications that detect temporary workers staying for extended periods after an emergency response and notify the administrator of an alarm, etc., form a structure that can respond meticulously and in real-time to a variety of on-site needs. Specifically, the entire system is orchestrated by a meta-AI model that monitors and controls each subsystem. The system has a dynamically expandable containerized microservice cluster infrastructure. This processing unit periodically collects health check data from each subsystem. This processing unit summarizes the collected data into a time-series tensor. This meta-AI model receives health check data from each subsystem as input data. This input data includes, for example, the percentage of CPU utilization. This input data includes, for example, the number of bytes of free memory. This input data includes, for example, network bandwidth consumption. This meta-AI model generates a resource optimization plan for the entire system as output data. This output data includes, for example, the number of containers that should be additionally started. This output data includes, for example, the IDs of nodes that should be scaled down. This output data includes, for example, a list of services that need to be restarted. This system issues commands to the orchestration tool based on the output plan. This system automatically scales up resources when traffic increases sharply. This system redirects traffic to healthy nodes when a failure occurs. This model uses past system operation logs and failure history for reinforcement learning. This learning process sets a reward function to maximize the availability of the entire system. This learning process uses policy gradients to update resource allocation strategies. This architecture significantly improves the fault tolerance of the entire system. This architecture enables the system to self-repair and self-optimize without administrator intervention. This architecture strictly adheres to service level agreements while minimizing operational costs.This technology can be applied to infrastructure management of cloud platforms requiring 24 / 7 / 365 uninterrupted operation. It can also be applied to disaster recovery systems that ensure business continuity even during large-scale disasters. This system is not merely a collection of functions, but rather a self-contained system that functions organically and collaboratively. It offers ultimate flexibility and robustness compared to previous static system architectures.

[0077] (System Composition) The system involved in this implementation includes a User Information Management Department, a Permission Granting Management Department, an Audit Record Management Department, a Generative AI Processing Department, an API Linkage Department, a User Interface Department, and an Authentication Processing Department. The User Information Management Department centrally manages all users entering and exiting the data center in the database, including maintenance technicians, maintenance workers, external contractors, security officers, cleaning staff, equipment inspectors, temporary workers, and audit officers. This includes a wide variety of attribute information such as name, ID, role, company affiliation, employment status, contact information, shift schedule, responsible area, past entry / exit history, qualification information, emergency contact information, health status, work plan, responsible equipment, access history, and access gate usage history. For example, when an external contractor enters the server room only during a specific period, detailed registration of that period, responsible area, work content, scheduled work time, work supervisor, required qualifications, availability of work instructions, and even emergency contact information can be recorded. Similarly, when an equipment inspector temporarily enters the data center for an annual inspection, the scheduled inspection date, responsible area, required equipment list, inspection content, and deadline for submitting the inspection report can be recorded. For example, when cleaning staff enter a specific area at fixed times each week, this can be registered as a regular schedule in the work schedule and managed in conjunction with their entry and exit history. Similarly, when temporary workers enter for emergency response, the system can record the reason for the emergency, the content of the response, the start and end times, the person in charge, and the reporting obligations afterward. Specifically, this system employs an advanced data management architecture that links distributed hash tables with a graph database. The system features an in-memory data grid for ultra-low latency data access. This processing unit unfolds the complex relationships between users and resources as a graph structure in memory. This processing unit asynchronously invalidates and rebuilds the cache during data updates. This generative AI processing unit receives user-resource relational data as input data. This input data may include, for example, pairs of user IDs and resource IDs. This input data may include, for example, tags indicating access permission types. This input data may include, for example, valid timestamps indicating expiration dates. This generative AI processing unit generates update queries for the relational graph as output data. This output data may include, for example, SQL commands for appending nodes. This output data may include, for example, updated edge weight values. This output data may contain unwanted relational deletion flags. The User Information Management Department uses the output queries to execute database transactions. The User Information Management Department broadcasts updated results to all nodes to maintain consistency. The User Information Management Department dynamically recalculates data sharding keys to prevent load skew. This model uses past query execution history to learn and optimize execution plans. This learning process adjusts the index structure to minimize query response time. This learning process uses reinforcement learning to explore optimal caching strategies.This structure retrieves and updates massive relational graphs consisting of millions of nodes and edges in milliseconds. It offers linear scalability even with increasing data volume. This structure instantly resolves complex permission dependencies and eliminates access latency. This technology can be applied to access control infrastructure for social network services with hundreds of millions of users. This technology can be applied to platforms for authentication and permission management of massive IoT devices. This system efficiently handles complex many-to-many relationships that are difficult to manage in existing relational databases. This system achieves intuitive data modeling through graph operations in high-dimensional space.

[0078] The access control department leverages generative AI to comprehensively analyze each user's role, affiliated company, business content, work plan, qualification information, past access performance, scheduling information, responsible area, health status, work supervisor, work content, and emergency response history. It automatically assigns the minimum and optimal logical and physical access permissions. For example, it grants maintenance technicians access to the server room and monitoring room, as well as operational permissions to the server management system; external contractors are granted access to the server room only during work hours and temporary configuration change permissions; security managers are granted access to all areas and full access to the monitoring system; cleaning staff are granted physical access only to the cleaning area; equipment inspectors are granted physical access only to their designated inspection areas and access to the equipment management system; and audit managers are granted access to all areas and log access only during audits. For example, AI learns from past access history and anomaly detection history to instantly detect unusual actions such as "external contractors attempting to enter power rooms that are not normally accessed," "maintenance technicians attempting to enter server rooms late at night," "cleaning personnel attempting to enter areas outside their assigned areas," "equipment inspectors attempting to enter areas outside their scheduled inspection areas," and "temporary workers remaining for extended periods after an emergency response has ended," and alerts the administrator accordingly. For example, based on qualification information, access to specific areas or systems can be granted only to users with specific advanced security qualifications. For example, physical access can be temporarily suspended and the administrator notified if a system is deemed to be in poor health. For example, operator access can be temporarily restricted if the person in charge of the operation is absent. Specifically, this permission granting management department a hybrid inference architecture integrating a rule-based inference engine and neural networks. This system has a dedicated tensor processing unit for high-speed inference processing. This processing unit vectorizes and combines static security policies and dynamic context information. This processing unit normalizes the combined vectors. This generative AI processing unit receives vectors of static rules and dynamic context as input data. This input data may include, for example, a text vector of the company's security policy. This input data may include, for example, a score of current threat intelligence. This input data may include, for example, the user's recent activity history data. This generative AI processing unit generates the final permission decision as output data. This output data may include, for example, a true / false value for access permission. This output data may include, for example, a list of applicable rule IDs. This output data may include, for example, a flag indicating the necessity of anomaly handling. This permission grants the management department the authority to immediately reflect the output decision in the access control list. This permission grants the management department the authority to record the rule IDs used as the basis for the decision in the audit log. This permission grants the management department the authority to escalate to a human administrator if the anomaly handling flag is true.This model uses past successful and unsuccessful examples of permission granting for supervised learning. This learning process simultaneously optimizes the branching conditions of the decision tree and the weights of the neural network. This learning process maximizes prediction accuracy while maintaining the interpretability of the reasoning. This structure achieves flexible permission granting while meeting strict compliance requirements. This structure brings the technical effect of providing clear and easily black-boxed AI judgment criteria. This structure minimizes damage by dynamically reducing permissions even for unknown threats. This technology can be applied to strict access control systems in military facilities. This technology can be applied to the management of electronic medical record access permissions in medical institutions. This system utilizes both rules set by human administrators and patterns discovered by AI to make optimal judgments. This system instantly interprets complex contexts through vector operations in high-dimensional space.

[0079] The Audit Log Management Department meticulously records all user account operations, permission changes, physical room entry and exit records, logical system access records, data synchronization history with external systems via API, terminal information, and time information. For example, when maintenance technicians enter the server room, log into the server management system, make configuration changes, and implement fault handling, this series of physical and logical access and operation history records are recorded chronologically. Similarly, if an external contractor attempts to enter a room outside of work hours, the attempt record and authentication failure history at the turnstile are also meticulously retained. Furthermore, if an equipment inspector submits an inspection report after an inspection, this submission history is also saved as an audit log. For significant permission changes and account deletion operations, multi-factor authentication by the administrator is mandatory, and electronic signatures are assigned to the operation records to prevent subsequent tampering or denial. Audit logs are backed up regularly and also have archiving capabilities for long-term storage. For example, all permission change history, room entry and exit history, operation history, and anomaly detection history from the past five years are securely stored, enabling rapid retrieval and extraction as needed. Specifically, this audit log management unit installs a multi-task learning model that simultaneously optimizes anomaly detection and data compression. The system possesses a storage management infrastructure that automatically stratifies hot and cold data. This processing unit parses the continuously flowing raw audit log stream in real time. This processing unit determines the storage layer for the target log based on its importance. This generative AI processing unit receives the raw audit log stream as input data. This input data may include, for example, trace data of system calls. This input data may include, for example, header information of network packets. This input data may include, for example, error log text from applications. This generative AI processing unit generates compressed log data and anomaly flags as output data. This output data may include, for example, reversibly compressed binary data. This output data may include, for example, an index of log lines containing anomalies. This output data may include, for example, category scores indicating importance. This audit log management unit writes the output binary data to the designated storage layer. This audit log management unit sends a notification to the real-time dashboard when the anomaly flag is true. This audit log management unit asynchronously updates the retrieval index to maintain query response speed. This model uses a large amount of past log data for self-supervised learning. This learning process optimizes the trade-off between data reconstruction error and compression ratio. This learning process studies the latent space of a variational autoencoder. This architecture efficiently stores petabytes of log data while enabling instantaneous retrieval of necessary information. This architecture delivers a significant reduction in storage costs. This architecture completely prevents log tampering and deletion, meeting the requirements for evidence preservation. This technology can be applied to the long-term storage systems of telecommunications operators' massive communication logs. This technology can be applied to compliant archiving systems that store the complete transaction history of financial institutions.This system does not simply compress data; instead, it dynamically adjusts its storage strategy based on the semantic importance of logs. Furthermore, it discovers unknown anomaly patterns in logs through feature extraction in a high-dimensional space.

[0080] The generative AI processing unit comprehensively analyzes diverse data, including user information, permission granting history, audit logs, system utilization, event information from external systems, health status, work plans, emergency response history, equipment inspection history, and cleaning history. Using anomaly detection algorithms, it detects unusual access and permission changes in real time. For example, it automatically identifies anomalies such as users who don't normally access important areas late at night, external contractors attempting to enter the power room, unauthorized use of accounts belonging to former employees, cleaning staff entering areas outside their assigned areas, equipment inspectors entering areas outside their scheduled inspection areas, and temporary workers remaining in the area for extended periods after an emergency response. It then notifies administrators of these anomalies. Furthermore, by analyzing administrator operation history and past audit logs, it identifies tendencies towards incorrect permission settings and personalization, automatically reporting risks such as "frequent permission granting errors caused by specific administrators," "frequent suspicious access by specific external contractors," and "concentrated abnormal access in specific areas." It can also propose corrective measures and educational initiatives. Finally, AI analysis of user movement and permission transfer history across multiple data center sites can automatically assess and adjust the consistency of security policies across these sites. For example, AI analysis of equipment inspection and cleaning history can detect omissions and illegal operations, and report to the administrator. Specifically, this generative AI processing unit is equipped with a Transformer-based neural network model with a self-attention mechanism. This generative AI processing unit generates a multi-dimensional numerical vector containing user attribute information, access time, location coordinates, and biometric authentication score as input data. This generative AI processing unit, for example, receives a 512-dimensional time series tensor as input data. This generative AI processing unit performs normalization and missing value imputation on the input data. As an example of input data, this generative AI processing unit uses a vector containing user identifier, access request time, target area identifier, and the number of accesses in the last 24 hours. As another example of input data, this generative AI processing unit uses a tensor that converts natural language text representing the administrator's operation history into word embedding vectors. This generative AI processing unit operates on these input tensors in high-dimensional space to calculate anomaly scores. As output data, this generative AI processing unit outputs anomaly probability scores with continuous values ​​ranging from 0 to 1. As one example of output data, this generative AI processing unit outputs a value indicating an anomaly probability of 0.95 for a specific access request. As another example of output, this generative AI processing unit outputs a classification label and its confidence level indicating the probability of incorrect permission settings. If the output anomaly probability score exceeds a preset threshold, this generative AI processing unit generates an access blocking command for further processing. This generative AI processing unit then sends the generated command to the access control module.This generative AI processing unit, as an anomaly detection algorithm, employs a method for calculating the reconstruction error using an autoencoder. It pre-trains the autoencoder on a dataset of normal access patterns. The unit uses mean squared error as the error function during learning. It optimizes the network's weight parameters using gradient descent. The unit calculates the difference between the input access data and the reconstructed data. If this difference exceeds a predetermined benchmark value, it is considered an anomaly. This process allows for the detection of minute anomalies that are difficult to detect instantly by human visual surveillance. As a result, it provides a preventative effect against security incidents. Applications include physical security management in data centers and access monitoring of critical systems in financial institutions. When evaluating the consistency of security policies across branches, it represents the policy settings of each branch as graph-structured data. It uses a graph neural network to extract the differences in settings between branches. Based on these extracted differences, it recommends the optimal policy settings. This generative AI processing unit sends the recommendation results to the administrator's terminal. This significantly improves the computational efficiency of security management in large-scale distributed systems.

[0081] The API Linkage Department integrates data with external systems such as HR, Access Control, Equipment Management, Cloud Services, Local Systems, Health Management, and Emergency Notification Systems via APIs. For example, it receives onboarding, transfer, and departure events from the HR system; real-time room entry and exit events from the Access Control system; inspection schedule information from the Equipment Management system; health status information from the Health Management system; and emergency response information from the Emergency Notification system, reflecting this information immediately in the User Information Management Department and the Access Control Department. For instance, it synchronizes authentication and permission information with external systems such as cloud storage services and surveillance camera systems to achieve single sign-on and unified access control. For example, upon receiving inspection schedule information from the Equipment Management system, it automatically adjusts the access permissions of corresponding inspectors. For example, upon receiving contract termination information from an external contractor, it immediately revokes all physical and logical access permissions and saves the entire operation history as an audit log. For example, upon receiving abnormal health information from the Health Management system, it temporarily suspends the access permissions of the corresponding user and notifies the administrator. Specifically, this API Linkage Department uses asynchronous communication protocols to maintain constant connections with multiple external systems. The API Linkage Department converts received event data into a standardized JSON data structure. This API linkage unit stores the converted data in the message queue system. It then retrieves data sequentially from the message queue and performs data cleaning. During data cleaning, it removes unwanted characters and standardizes date formats. Finally, it writes the processed data to the distributed database of the user information management department. This process is executed with millisecond delays. Mutual authentication is used in communication with external systems. The communication path is protected by an encryption protocol. Receiving health anomaly information from the health management system, this API linkage unit obtains time-series numerical arrays of heart rate and body temperature. It expands these time-series numerical arrays into a one-dimensional tensor in memory. This tensor is then input into the anomaly detection module. As output from the anomaly detection module, this API linkage unit receives a control flag indicating a temporary halt to access permissions. Based on this control flag, it sends a permission revocation command to the permission granting and management department. This mechanism minimizes the latency of data synchronization between systems. This API integration effectively mitigates the risk of unauthorized account placement by departing employees and those whose contracts have ended. It is applied in use cases such as employee access control in large-scale healthcare facilities and access management for external contractors in manufacturing plants. This API integration utilizes open authentication and authorization protocols in its synchronization with cloud storage services. It dynamically generates access tokens and sends them to external systems. This API integration strictly manages the expiration dates of these tokens.This API linkage automatically discards access requests using expired tokens. This reduces network traffic while maintaining a high level of security.

[0082] The user interface provides screens for users and administrators to confirm their account information and permissions, and to submit and approve applications. For example, when an external contractor requests an extension of the work period, the application information is sent from the terminal. The generative AI on the server side analyzes the application content, automatically determines whether the required permissions are appropriate, and returns approval or rejection feedback to the terminal. For example, when an equipment inspector obtains new qualifications, the information is entered from the terminal, and the AI ​​automatically grants access permissions to the corresponding equipment. Administrators can view audit logs and AI-based risk alerts from the terminal and instruct additional investigations and corrective actions as needed. For example, when temporary workers apply to leave the site after an emergency response, the AI ​​automatically assesses the response content and stay time, and notifies the administrator of any suspicious points. Specifically, this user interface has a dynamic webpage generation engine. This user interface sends screen data consisting of HTML and JavaScript to the user's terminal. This user interface forwards input data from the terminal to the server via asynchronous communication. When receiving application information, this user interface transfers the data to the generative AI processing unit containing a natural language processing model. This generative AI processing unit receives a dataset as input data containing a text string describing the reasons for the application, the applicant's identifier, and the identifier of the requested permissions. The generative AI processing unit tokenizes the input text and converts it into word embedding vectors. As an example of input data, the generative AI processing unit uses the text vector "I want to extend the operation time by two hours for a firmware update of the network equipment." As another example of input data, the generative AI processing unit uses structured data containing the name of the newly acquired qualification and the date of acquisition. The generative AI processing unit feeds this input data into a multilayer perceptron and calculates an appropriateness score. As output data, the generative AI processing unit outputs an appropriateness probability score ranging from 0 to 1. As an example of output data, the generative AI processing unit outputs an appropriateness score of 0.88 for the extension application. As another example of output data, the generative AI processing unit outputs a binary flag indicating whether or not permissions are granted based on the qualification information. If the output score exceeds a predetermined threshold, the generative AI processing unit generates an automatic approval command as subsequent processing. This generative AI processing unit generates an escalation request to the administrator when the score is below a threshold. The user interface unit receives this decision and instantly displays an approval or rejection message on the terminal screen. This user interface unit reduces the manual approval process for administrators. As a result, this user interface unit delivers a significant improvement in the processing speed of the permission granting process. This user interface unit is applied to use cases such as global IT support desks in multinational corporations and permission management for personnel in large-scale construction sites. When displaying risk alerts, this user interface unit uses a data visualization library to generate charts and heatmaps.This user interface is designed to allow administrators to intuitively identify abnormal areas. This user interface thereby improves the efficiency of human-computer interaction.

[0083] The authentication processing department implements identity authentication for users and administrators accessing the system, combining multi-factor authentication such as password authentication, one-time password authentication, biometric authentication, terminal authentication, facial recognition, and voice authentication to prevent impersonation and unauthorized access. For example, when an administrator makes important permission changes, one-time password authentication, fingerprint authentication, and facial recognition are required on the terminal, and the authentication result is sent to the server to authorize the operation. Authentication history records are sent to the audit record management department, and all authentication events are recorded. For example, combining facial recognition and IC card authentication at access gates can further strengthen identity verification. Specifically, this authentication processing department has an authentication orchestrator that centrally manages multiple authentication modules. When performing facial recognition, this authentication processing department receives two-dimensional image data acquired from the terminal's camera as input. This authentication processing department performs preprocessing on the input image data, including face region segmentation and normalization. This authentication processing department converts the preprocessed image data into a three-dimensional image tensor. This authentication processing department inputs this image tensor into a convolutional neural network. As an example of input data, this authentication processing department uses a 224-pixel square RGB image tensor. As another input example, this authentication processing unit uses a two-dimensional array of data containing Mel-frequency cepstral coefficients for voice authentication. This authentication processing unit uses a convolutional neural network to extract a 512-dimensional feature vector from the input image. This authentication processing unit calculates the cosine similarity between the extracted feature vector and a pre-registered reference vector in the database. As output data, this authentication processing unit outputs a similarity score ranging from 0 to 1. As an example of output data, this authentication processing unit outputs a similarity score of 0.97 for face authentication. As another output example, this authentication processing unit outputs a true / false value indicating successful voice authentication. If the output similarity score exceeds a set authentication threshold, this authentication processing unit generates an authentication success token for further processing. This authentication processing unit sends the generated token to the session management module. This authentication processing unit installs a risk-based authentication algorithm that dynamically determines the combination of multi-factor authentication. This authentication processing unit calculates a risk score based on the IP address and time period of the access source. If the risk score is high, this authentication processing unit sends a control signal to the terminal requesting additional biometric authentication. This authentication processing unit significantly improves authentication accuracy compared to existing systems that rely on static password authentication through this mechanism. As a result, it minimizes the risk of information leakage due to unauthorized access. This authentication processing unit is applied in use cases such as confidential information management systems in government agencies and core banking systems in financial institutions. This authentication processing unit uses a cryptographic hash function to hash authentication history data. The hashed data is then asynchronously sent to the audit log management department. This authentication processing unit thus achieves reliable logging without reducing the response speed of authentication processing.

[0084] Specific examples of the prompts required for generative AI to read when implementing this invention include: "granting external contractors access to the server room only during the operation period and automatically revoking it after the period ends," "extracting the access history of users who entered the server room late at night and detecting abnormal patterns," "analyzing the past entry and exit history of users with access to the power room and listing actions that are different from usual," "immediately invalidating the logical and physical access rights of departed users and saving all operation history as audit logs," "analyzing the tendency of administrators to grant permissions and extracting cases with high risk of errors and personalization," "detecting equipment inspectors entering areas outside their designated inspection area and immediately notifying the administrator," "automatically temporarily suspending access rights when health is abnormal and re-granting them upon recovery," and "detecting temporary workers staying for extended periods after an emergency response and notifying the administrator of an alarm," etc. Specifically, this system uses these prompts as input data to the natural language processing model. The system segments the strings constituting the prompts into word units and converts them into a one-dimensional array of token IDs. As one example of input data, this system uses an array of tokens containing an instruction to "grant external contractors access to the server room only during the operation period, and automatically revoke it after the period ends." As another example of input data, this system incorporates access history data extracted from system logs into a JSON-formatted text string into the prompt. This system feeds this input data into the encoder layer of a large language model. Internally, the system uses a self-attention mechanism to calculate dependencies between words within the prompt. As output data, this system generates a sequence of specific system commands to be executed. As another example of output data, this system outputs API call code granting permissions to a specific user ID. As another example of output data, this system outputs a structured report summarizing anomaly detection results in JSON format. This system passes the output command sequence to the syntax parsing module, converting it into an executable format. This system sends the converted commands to various control modules to execute actual permission changes and alarm notifications. In generating the prompt text, this system combines template-based generation techniques with dynamic parameter embedding. This system automatically selects the optimal prompt template based on the system's state and the type of event. This system embeds user ID and time information into the selected template to construct the final prompt text. This process eliminates the need for humans to manually input complex settings commands. As a result, this system improves the automation level of system operation and prevents security incidents caused by configuration errors. This system is applicable to use cases such as automated operation and management of cloud infrastructure and unmanned operation of large-scale data centers. In the learning of large-scale language models, this system uses a professional corpus related to system operation for fine-tuning.This system thus gains the ability to generate application commands with higher accuracy than a typical language model.

[0085] (Implementation steps) Step 1: Detailed User Information Registration and External Linkage. For all users entering and exiting the data center, a variety of attribute information, including names, IDs, roles, affiliated companies, employment status, contact information, shift information, responsible areas, past entry / exit history, qualification information, emergency contact information, health status, work plans, responsible equipment, access history, and gate usage history, is uniformly registered and managed in the User Information Management Department. This information is integrated with the Human Resources System, Access Management System, Equipment Management System, Health Management System, and Emergency Notification System via API. The system receives real-time event information such as employee onboarding, transfers, departures, work plans, health status, and emergency responses, and automatically synchronizes and updates user information. Specifically, this User Information Management Department utilizes a distributed relational database system. The department stores each user's attribute information in a standardized table structure. The department also receives event data from external systems as input. As one example of input data, the User Information Management Department uses XML format data containing the name, employee number, and assigned department of a new employee sent from the Human Resources system. As another example of input data, the User Information Management Department uses JSON format data containing the gate passage time and IC card identifier sent from the Access Management System. The User Information Management Department parses the received data and maps it to its internal data schema. The User Information Management Department writes the mapped data to the corresponding record in the database. During this write process, the User Information Management Department performs transaction management to ensure data consistency. When the data update is complete, the User Information Management Department generates an update completion notification event as output data. As an example of output data, the User Information Management Department outputs a message indicating that the status of a specific user ID has changed from "employed" to "departed." As another example of output data, the User Information Management Department outputs an alarm flag indicating that health status parameters deviate from the normal range. The User Information Management Department sends the generated event to the Permissions Management Department via a message broker. As a subsequent processing step, the Permissions Management Department uses this event as a trigger to recalculate permissions. To improve database retrieval performance, this User Information Management Department employs a memory caching system. Frequently accessed user information is kept in the cache memory. This reduces the query load on the database. As a result, this User Information Management Department maintains millisecond-level response times even when processing user information for tens of thousands of users. This User Information Management Department is applied in use cases such as resident ID management infrastructure in large-scale smart cities and unified global human resources information systems for multinational corporations. This User Information Management Department asynchronously performs data backup processing, enabling rapid recovery in the event of a failure.

[0086] Step 2: Dynamic Permission Assignment and Adjustment Based on Generative AI. In the permission assignment management department, generative AI is utilized to comprehensively analyze each user's role, company affiliation, business content, work plan, qualification information, past access performance, scheduling information, responsible area, health status, work supervisor, work content, and emergency response history. This automatically assigns the minimum and optimal logical and physical access permissions, and adjusts permissions in real time based on changes during the work period, responsible area, health status, qualification status, and emergency response status. Specific examples of prompts read by the generative AI include: "Grant access to the server room of external contractors only during the work period, and automatically revoke it after the period ends," "Automatically suspend access permissions when health is abnormal, and re-grant them upon recovery," "Detect equipment inspectors entering unplanned areas and immediately notify the administrator," and "Dynamically control access permissions for specific areas and systems based on qualification information." Specifically, this permission assignment management department installs a deep reinforcement learning model as the generative AI. The department inputs user attribute information and context information as state vectors. This permission grants the management department the following privileges as input data: A 128-dimensional numerical vector containing the user's job ID, current time, importance level of the target system, and recent health status score. Another example of input data is an array of timestamps representing start and end times extracted from the work schedule. This permission grants the management department this privilege to feed this input data into a multi-layer neural network to calculate the probability distribution of whether each privilege should be granted. The output data represents a vector indicating the probability of granting all privileges within the system. Another example of output data is an array containing a probability of 0.99 for granting physical access to the server room and 0.01 for granting database administrator privileges. Another example of output data is a natural language text string explaining the reason for the privilege change. This permission grants the management department the privilege to extract outputs with probabilities exceeding a predetermined threshold, as part of subsequent processing to generate an update command for the access control list. This permission grants the management department the privilege to send the generated command to the access control gateway of each system. This permission grants the management department the power to use a reward function in reinforcement learning, assigning a negative reward to the occurrence rate of security incidents and a positive reward to the smoothness of business execution. This permission also allows the management department to continuously update the model's weights using past access logs and event history. This process enables the management department to achieve flexible, situation-based access control that static, rule-based access management cannot handle. As a result, this permission maximizes business efficiency while eliminating security risks caused by excessive permission granting. This permission can be applied to use cases such as strict access control systems for military facilities and access management in financial institution trading halls.This authority grants the management department the power to record the history of authority adjustments in the blockchain network, ensuring transparency in auditing.

[0087] Step 3: Real-time modification and revocation of permissions based on event triggers. Through the API linkage department, the system receives real-time event information from the HR system, access control system, equipment management system, health management system, and emergency notification system, including employee onboarding, transfers, departures, end of work periods, health abnormalities, and emergency response completion. Permissions are immediately assigned, modified, or revoked for the corresponding users, and all permission change history is recorded in the audit record management department. Examples include the termination of contracts for external contractors, the end of inspection work for equipment inspectors, the suspension of access permissions during health abnormalities, and the revocation of permissions for temporary workers after an emergency response. Specifically, this system adopts an event-driven architecture. The system has an event router module that branches processing based on the type of received event. The system receives a structured payload as input data, containing the source system identifier, event category code, target user ID, and occurrence time of the event. As an example of input data, this system uses a JSON object containing an event code indicating "departure" issued from the HR system and the corresponding employee's employee number. As another example of input data, this system uses alarm data indicating "abnormal body temperature" sent from the health management system. This system feeds this input data into the event processing engine. Within the event processing engine, the system infers the required actions using a predefined rule engine or a learned decision tree model. As output data, the system generates a series of API requests to execute permission changes. As another example of output data, the system outputs an account invalidation command for Active Directory. As yet another example, the system outputs an access permission cancellation command for a physical electronic lock system. The system sends the generated commands in parallel to the target systems as subsequent processing. The system awaits completion responses from each system to confirm that all processing has been successful. The system formats the processing results into an audit log and sends it to the audit log management department. This system completes this series of processes within hundreds of milliseconds from receiving the event. This completely eliminates the time lag associated with manual permission changes by human administrators. As a result, the system effectively prevents malicious data tampering after leaving the company and prevents individuals with health issues from performing hazardous work. This system is applicable to use cases such as personnel management systems in nuclear power plants and access control in biosafety-critical research institutes. In the event that communication with the target system fails due to network failures or other reasons, this system uses an exponential backoff algorithm to automatically perform retry processing.

[0088] Step 4: Detailed Collection and Tamper-Proofing of Audit Logs. The audit log management department records all user account operations, permission changes, physical room entry and exit records, logical system access records, data synchronization history with external systems via API, terminal information, and time information in detail. For significant permission changes or account deletion operations, multi-factor authentication and electronic signatures are mandatory to prevent tampering or denial of audit logs. Audit logs are backed up regularly and also have archiving capabilities for long-term storage. Specifically, this audit log management department is built on an append-only distributed ledger technology infrastructure. This audit log management department receives log data from various modules as input. As an example of input data, this audit log management department uses log entries containing the operator's user ID, the URI of the target resource, the HTTP method executed, and the current timestamp of the operation date. As another example of input, this audit log management department uses the administrator's biometric authentication result and electronic signature data based on a public-key cryptography system accompanying the permission change operation. This audit log management department calculates the hash value of the input log data using the SHA-256 algorithm. This audit log management system generates a new hash value by combining the hash value of the immediately preceding log entry with the current log data, thus forming a cryptographic chain structure. As output data, this audit log management system outputs log records that are determined as new blocks on the blockchain. As another example of output data, this audit log management system outputs transaction IDs that guarantee immutability. As yet another example of output, in response to log retrieval queries from auditors, this audit log management system outputs a list of log data sorted chronologically. This audit log management system asynchronously transmits the determined log data to cold storage for subsequent processing and archives it. In the verification of electronic signatures, this audit log management system uses certificates obtained from certification authorities to mathematically prove the legitimacy of the signature. Through this process, this audit log management system ensures that even insiders or privileged administrators cannot tamper with or delete past operation history records. As a result, this audit log management system delivers a significant improvement in system auditability and rigorous compliance requirements. This audit log management system is applied to use cases such as operation log management in the financial accounting systems of listed companies and access record management for electronic medical records in medical institutions. In order to optimize the storage capacity of log data, the Audit Record Management Department automatically compresses data after a certain period of time using a compression algorithm to reduce storage costs.

[0089] Step 5: Anomaly Detection and Risk Alerts Based on Generative AI. In the generative AI processing unit, a variety of data are comprehensively analyzed, including user information, permission granting history, audit logs, system utilization, event information from external systems, health status, work plans, emergency response history, equipment inspection history, and cleaning history. Anomaly detection algorithms are used to detect abnormal access and permission changes, entry into unauthorized areas, suspicious behavior during late-night hours, access attempts by departed users, and prolonged stays by temporary staff. Risk notifications and corrective action proposals are then sent to the administrator. Specific examples of prompts read by the generative AI include: "Extract the access history of users entering the server room during late-night hours and detect abnormal patterns," "Immediately invalidate the logical and physical access permissions of departed users and save all operation history as audit logs," "Detect prolonged stays by temporary staff after emergency response and notify the administrator of the alert," and "Analyze the administrator's permission granting tendencies and extract cases with high risk of errors and personalization." Specifically, this generative AI processing unit is equipped with a Long Short-Term Memory (LSTM) network specialized for time-series data analysis. This generative AI processing unit integrates diverse log data collected from the entire system, generating feature tensors at regular time intervals as input data. One example of input data is a multidimensional vector containing elements such as the number of times users entered each area in the past hour, the number of failed login attempts, and the number of permission change operations. Another example is a tensor encoding a specific user's behavioral history over the past month as sequential data. This generative AI processing unit feeds this input data into a Long Short-Term Memory (LSTM) network to calculate the degree to which the current behavioral pattern deviates from past normal patterns. As output data, this generative AI processing unit outputs an array of anomaly scores for each user or system component. One example of output data is an anomaly score of 0.92 indicating that the temporary worker's dwell time significantly exceeded the predicted value. Another example of output data is a list of natural language text explaining the cause of the anomaly and recommended corrective measures. If the output anomaly score exceeds a warning threshold, this generative AI processing unit initiates an alarm generation module as a subsequent processing step. This alert generation module sends push notifications to the administrator's smartphone. In the learning process of the anomaly detection model, this generative AI processing unit combines supervised learning, which uses data from past security incidents as teacher data, with unsupervised learning, which uses only normal data. Through this hybrid learning approach, this generative AI processing unit can not only detect known attack patterns with high accuracy, but also detect unknown zero-day attacks and signs of internal violations with high accuracy.As a result, this generative AI processing unit achieves high levels of automation and sophistication in security surveillance, significantly reducing incident response time. It is applied in use cases such as national-scale critical infrastructure protection systems and monitoring unauthorized access to large-scale EC websites. To reduce false detections, this generative AI processing unit provides a mechanism for administrators to provide feedback on alert inputs, which is then used to relearn the model.

[0090] Step 6: Application, Approval, and Information Update via the User Interface. The user interface provides a screen for users or administrators to confirm their account information and permissions, and to submit applications or approvals. Examples include external contractors applying for work extensions, equipment inspectors applying for new qualifications, and temporary workers applying to leave after an emergency response. Application information is sent from the terminal, and the server-side generative AI parses the application content, automatically determines the appropriateness of the required permissions, and returns approval or rejection feedback to the terminal. Specifically, this user interface uses a single-page application architecture, running at high speed on the user's terminal. This user interface detects user input and constructs the input data into a JSON-formatted payload. One example of input data used by this user interface is a data object containing text describing the reason for the work extension, the desired extension period, and the current work progress rate. Another example of input data used by this user interface is multi-part form data containing an image of the uploaded qualification certificate and the name of that qualification. This user interface sends this input data to the server-side generative AI processing unit via a RESTful API. This generative AI processing unit applies a natural language processing model to text data and an optical character recognition model to image data to extract information. Based on the extracted information, this generative AI processing unit compares and verifies the application content with the current system status. As output data, this generative AI processing unit outputs an approval probability score for the application and text explaining the reasons for the decision. As an example of output data, this generative AI processing unit outputs an approval probability of 0.95 for a job extension application and the reason text "It is appropriate because it is consistent with the average time required for similar jobs in the past." As another example of output data, this generative AI processing unit outputs a warning sign indicating the possibility of certificate forgery. If the approval probability exceeds a threshold, this generative AI processing unit sends a permission update command to the permission granting management department as a follow-up process. This user interface unit receives responses from the server and dynamically updates the status display on the screen. Through this processing, this user interface unit completely digitizes the complex approval workflow based on paper applications or email. As a result, this user interface unit delivers a technological effect that drastically reduces the delivery time from application to permission granting from several days to several seconds. This user interface is designed for use cases such as temporary requests for electronic medical record access by doctors in large-scale hospitals and requests for access to research equipment by university students. To operate even in environments with unstable communication lines, this user interface features the ability to temporarily store request data in local memory when offline and automatically send it back when communication is restored.

[0091] Step 7: Enhance security through multi-factor authentication. In the authentication processing unit, user or administrator authentication is implemented when accessing the system. This combines multiple factors such as password authentication, one-time password authentication, biometric authentication, terminal authentication, facial recognition, and voice authentication to prevent impersonation and unauthorized access. Authentication history records are sent to the audit record management department to record all authentication events. Specifically, this authentication processing unit functions as an authentication gateway based on a zero-trust security model. Upon receiving an access request from a user, this authentication processing unit collects authentication context data. As an example of input data, this authentication processing unit uses a dataset containing the hash value of the user's entered password, the terminal's MAC address, and latitude and longitude coordinates obtained from GPS. As another example of input data, this authentication processing unit uses a biometric vector obtained from the fingerprint sensor of a smartphone terminal. This authentication processing unit inputs this input data into the authentication evaluation engine. This authentication evaluation engine contains a machine learning-based behavioral analysis model. This authentication evaluation engine compares the input context data with the user's past standard behavioral profiles. As output data, this authentication evaluation engine outputs a risk score for the current access request in the range of 0 to 100. As one example of output data, this authentication assessment engine outputs a risk score of 85 for access requests from overseas IP addresses that are different from usual. As another example, it outputs a risk score of 5 for access from a legitimate company intranet. Based on the output risk score, this authentication processing unit dynamically determines the number of authentication elements required for subsequent processing. In cases of high risk scores, this processing unit sends a prompt to the terminal requesting facial recognition in addition to a one-time password. This processing unit issues an access permission token only after all required authentication elements have been cleared. This authentication processing unit optimizes the trade-off between convenience and security through this dynamic multi-factor authentication mechanism. As a result, this authentication processing unit provides a physical and logical technical effect in preventing account hijacking due to password list attacks or phishing scams. This authentication processing unit is applied to use cases such as customer login systems for cryptocurrency exchanges and VPN access control in remote office environments for enterprises. This authentication processing unit encrypts all metadata collected during the authentication process and transmits it to the audit log management department via a secure communication channel.

[0092] (specific use case) For example, when an external contractor performs equipment replacement work in a server room within a data center, the User Information Management Department meticulously registers the contractor's name, ID, company affiliation, scheduled work period, responsible area, necessary qualifications, emergency contact information, work supervisor, past access history, health status, and work content. This information is automatically synchronized with the HR system, access control system, and health management system via the API linkage department before the work begins. The Permissions Management Department uses generative AI to automatically assign physical access permissions to the server room only during the work period, as well as logical access permissions required for equipment replacement (configuration changes, log viewing, fault response permissions, etc.). At the end of the work period, the AI ​​automatically revoks all permissions and records this history in detail in the Audit Records Management Department. During the work, the AI ​​monitors the passage time of access gates, the utilization status of operating terminals, and system access logs in real time. For example, if it detects late-night hours, entry into non-responsible areas, unusual operating patterns, or entry attempts due to health anomalies, it immediately notifies the administrator of an alert. When equipment inspectors temporarily enter the data center for annual inspections, they register the scheduled inspection date, responsible area, required equipment list, inspection content, and qualification information. The AI ​​only grants physical access to the corresponding area and viewing access to the equipment management system during the inspection period, automatically revoking permissions after the inspection. When cleaning staff enter rooms for regular cleaning, it links with scheduling information to grant physical access limited to the cleaning area. If they attempt to enter an area outside their responsibility, the AI ​​immediately detects this as an anomaly and notifies the administrator. When temporary workers enter for emergency response, they register the reason for the emergency response, the response content, and the start and end times. The AI ​​issues alerts when it detects prolonged stays after the response ends or entry into an area different from the response content. When administrators perform important operations such as granting permissions or deleting accounts, multi-factor authentication and electronic signatures are required, and all operation history is recorded in the audit record management department. The AI ​​analyzes all these historical records, automatically extracting errors in permission granting and personalization tendencies caused by specific administrators, concentrated abnormal access in specific areas, omissions in equipment inspections and cleaning operations, and illegal operations, proposing corrective measures and emphasizing the necessity of training for administrators. Specifically, this system handles these diverse use cases on a single, unified platform. In the case of equipment replacement work by external contractors, the system inputs work instruction data into a natural language processing model. As one example of input data, the system uses a composite dataset containing text data such as "Router replacement work on rack A, required time three hours" and the contractor's biometric authentication template. Another example of input data uses cleaning staff schedules and 3D coordinate data of the target cleaning area. The system parses this input data and calculates the scope of permissions that should be granted to each user.This system outputs differential access control lists for each user. As another example of output data, it outputs SSH connection permission rules for specific IP address ranges for external contractors. As yet another example, it outputs unlocking permission tokens for specific electronic locks for cleaning personnel. The system automatically executes permission revocation scripts as subsequent processing, triggered by timer events indicating the end of a work session. During work monitoring, the system utilizes a streaming data processing engine. It expands turnstile logs and command execution logs in memory in real time and inputs them into an anomaly detection model. As output of the anomaly detection model, the system calculates a risk score for each behavior. If the risk score exceeds a threshold, the system immediately sends a control signal to lock the physical turnstile. When parsing administrator operation history, the system uses a graph database to model the relationships between administrators and personnel to whom permissions are granted as a network graph. The system calculates network centrality indicators to quantitatively assess the personalization of permission granting operations concentrated on specific administrators. Through this processing, the system fully automates access management for diverse personnel in complex facilities. As a result, this system offers the technical benefits of preventing security vulnerabilities caused by human error and significantly reducing operational costs. It is applicable to use cases such as managing hyperscale data centers of large cloud providers and high-security buildings with multiple tenants. The system extracts corrective action proposals and displays them on a dashboard in order of priority.

[0093] Specific examples of prompts that can be read by generative AI include: "Granting external contractors access to the server room only during the operation period, and automatically revoking it after the period ends," "Extracting the access history of users entering the server room late at night to detect abnormal patterns," "Detecting equipment inspectors entering unplanned areas and immediately notifying the administrator," "Detecting attempts by cleaning staff to enter non-responsible areas and alerting the administrator," "Detecting temporary workers lingering for extended periods after an emergency response and alerting the administrator," "Immediately invalidating the logical and physical access rights of departed users and saving all operation history as audit logs," "Analyzing the administrator's tendency to grant permissions and extracting cases with high risk of errors and personalization," and "Detecting omissions and illegal operations during equipment inspections and cleaning and reporting them to the administrator." Specifically, this system has a prompt engineering module that dynamically generates these prompts and sends them to the inference API of a large language model. This system receives raw event log data occurring within the system and a set of predefined system constraint rules as input data. As an example of input data, this system uses spatial information data containing the cleaning staff's ID, the sensor coordinates of their current location, and polygon data of the permitted cleaning area. This system, as another input example, uses the difference between the equipment inspector's job completion report text and the actual equipment operation log. This system embeds this input data as context into the prompt text. This system uses a large language model to perform context-based inference. As output data, this system outputs structured data in JSON format defining the specific actions to be performed. As another example of output data, this system outputs a data object containing the alarm's importance level, the administrator ID of the notification target, and a summary description of the event. As yet another output example, this system outputs a list of equipment IDs identifying missed areas in the job. This system sends the output structured data to the task management system for further processing, automatically generating event work orders. In optimizing prompts, this system uses past inference results and administrator feedback as learning data to automatically fine-tune the prompt template. Through this processing, this system enables computers to perform complex situation judgments that are difficult to describe in rule-based systems through natural language instructions. As a result, this system achieves a significant improvement in system flexibility and adaptability, and enhances the defense against unknown security threats. This system is applied to use cases such as autonomous safety monitoring in smart factories and customer and employee behavior management in unmanned stores. To accelerate the inference processing of large language models, this system uses model quantization techniques to reduce parameter size, enabling low-latency inference on edge devices.

[0094] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires voice representing the user's input on the result of the specific processing. The control unit 46A sends the voice data representing the user's input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0095] Data generation model 58 is what is known as generative AI (Artificial Intelligence).

[0096] One example of data generation model 58 includes ChatGPT (registered trademark) (Internet search).<URL: https: / / openai.com / blog / chatgpt> Generative AI, etc.

[0097] The data generation model 58 is obtained by performing deep learning on a neural network. A prompt containing instructions is input to the data generation model 58, along with inference data such as speech data representing speech, text data representing text, and image data representing images (e.g., data from still images or moving images). The data generation model 58 infers the input inference data based on the instructions shown in the prompt and outputs the inference result in one or more data forms, such as speech data, text data, and image data. The data generation model 58 may include, for example, text generation AI, image generation AI, and multimodal generation AI. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization. The specific processing unit 290 performs the aforementioned specific processing while using the data generation model 58. The data generation model 58 may also be a fine-tuned model to output inference results from prompts without instructions; in this case, the data generation model 58 can output inference results from prompts without instructions. The data processing device 12, etc., includes multiple data generation models 58, including AI other than generative AI. AI other than generative AI includes, but is not limited to, linear regression, logistic regression, decision trees, random forests, support vector machines (SVM), k-means clustering, convolutional neural networks (CNN), recurrent neural networks (RNN), generative adversarial networks (GAN), or Naive Bayes. Furthermore, AI can also act as an AI agent. Additionally, when the processing described above is performed by AI, this processing may be partially or entirely performed by AI, but is not limited to this example. Moreover, processing performed by AI, including generative AI, can be replaced by rule-based processing, and vice versa.

[0098] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart device 14 or external devices, and the smart device 14 acquires or collects information required for processing from the data processing device 12 or external devices.

[0099] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart device 14.

[0100] [Second Implementation] Figure 3 An example of the configuration of the data processing system 210 according to the second embodiment is shown.

[0101] like Figure 3 As shown, the data processing system 210 includes a data processing device 12 and smart glasses 214. One example of the data processing device 12 is a server.

[0102] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0103] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, and communication I / F 44 are also connected to the bus 52.

[0104] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0105] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0106] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0107] Figure 4 An example of the main functions of the data processing device 12 and the smart glasses 214 is shown. Figure 4 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0108] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0109] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290. The specific processing unit 290 can use the emotion-specific model 59 to infer the user's emotions and perform specific processing based on the user's emotions. The emotion inference function using the emotion-specific model 59 (emotion-specific function) includes various inferences and predictions related to the user's emotions, but is not limited to this example. Furthermore, emotion inference and prediction also include, for example, emotion analysis (analysis).

[0110] In the smart glasses 214, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating as a control unit 46A based on the receive / output program 60 executed on the RAM 48. Furthermore, the smart glasses 214 may also have the same data generation model and emotion-specific model as the data generation model 58 and the emotion-specific model 59, and use these models to perform the same processing as the specific processing unit 290.

[0111] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the smart glasses 214. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal".

[0112] (Example 1) Since the specific processing flow is the same as that in Embodiment 1 described in the first embodiment above, the description is omitted.

[0113] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0114] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0115] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or external devices, and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or external devices.

[0116] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the smart glasses 214.

[0117] [Third Implementation] Figure 5 An example of the configuration of the data processing system 310 according to the third embodiment is shown.

[0118] like Figure 5 As shown, the data processing system 310 includes a data processing device 12 and a head-mounted terminal 314. An example of the data processing device 12 is a server.

[0119] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0120] The head-mounted terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, display 343, and communication I / F 44 are also connected to the bus 52.

[0121] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0122] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0123] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0124] Figure 6 An example of the main functions of the data processing device 12 and the head-mounted terminal 314 is shown. Figure 6 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0125] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0126] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the specific processing unit 290.

[0127] In the head-mounted terminal 314, the processor 46 performs the receiving and output processing.

[0128] The memory 50 stores a receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on the RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on the RAM 48.

[0129] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the head-mounted terminal 314. In the following description, the data processing device 12 will be referred to as the "server" and the head-mounted terminal 314 will be referred to as the "terminal".

[0130] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0131] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0132] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the head-mounted terminal 314. In the head-mounted terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user's input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0133] Furthermore, the processing performed by the aforementioned data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the head-mounted terminal 314, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the head-mounted terminal 314. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the head-mounted terminal 314 or external devices, and the head-mounted terminal 314 acquires or collects information required for processing from the data processing device 12 or external devices.

[0134] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the head-mounted terminal 314.

[0135] [Fourth Implementation] Figure 7 An example of the configuration of the data processing system 410 according to the fourth embodiment is shown.

[0136] like Figure 7 As shown, the data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[0137] The data processing apparatus 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" as understood in this disclosure. The computer 22 includes a processor 28, RAM 30, and memory 32. The processor 28, RAM 30, and memory 32 are connected to a bus 34. Furthermore, the database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. An example of the network 54 includes a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0138] Robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control object 443. Computer 36 includes a processor 46, RAM 48, and memory 50. The processor 46, RAM 48, and memory 50 are connected to a bus 52. Furthermore, the microphone 238, speaker 240, camera 42, control object 443, and communication I / F 44 are also connected to the bus 52.

[0139] Microphone 238 receives instructions from user 20 by receiving the sounds emitted by user 20. Microphone 238 captures the sounds emitted by user 20, converts the captured sounds into speech data, and outputs it to processor 46. Speaker 240 outputs sound according to the instructions from processor 46.

[0140] Camera 42 is a small digital camera equipped with an optical system such as a lens, aperture and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, used to capture the area around user 20 (e.g., the shooting range defined by an angle of view equivalent to the field of vision of an average healthy person).

[0141] Communication I / F 44 is connected to network 54. Communication I / F 44 and 26 are responsible for the transmission and reception of various information between processor 46 and processor 28 via network 54. The transmission and reception of various information between processor 46 and processor 28 using communication I / F 44 and 26 is performed in a secure state.

[0142] The controlled object 443 includes a display device, LEDs for the eyes, and motors for driving the arms, hands, and feet. The posture and movements of the robot 414 are controlled by controlling the motors for the arms, hands, and feet. Some of the robot 414's emotions can be expressed by controlling these motors. In addition, facial expressions of the robot 414 can also be expressed by controlling the illumination state of the LEDs for the robot 414's eyes.

[0143] Figure 8 An example of the main functions of the data processing device 12 and the robot 414 is shown. For example... Figure 8 As shown, in the data processing device 12, specific processing is performed by the processor 28. The memory 32 stores a specific processing program 56.

[0144] The specific processing program 56 is an example of a "program" as disclosed herein. The processor 28 reads the specific processing program 56 from the memory 32 and executes the read specific processing program 56 on the RAM 30. Specific processing is implemented by the processor 28 operating as a specific processing unit 290 according to the specific processing program 56 executed on the RAM 30.

[0145] The memory 32 stores a data generation model 58 and an emotion-specific model 59. The data generation model 58 and the emotion-specific model 59 are used by the special processing unit 290.

[0146] In robot 414, the processor 46 performs the receive / output processing. The memory 50 stores the receive / output program 60. The processor 46 reads the receive / output program 60 from the memory 50 and executes the read receive / output program 60 on RAM 48. The receive / output processing is implemented by the processor 46 operating the control unit 46A based on the receive / output program 60 executed on RAM 48.

[0147] Next, the specific processing performed by the specific processing unit 290 of the data processing device 12 will be described. Each part of the system described below is implemented by the data processing device 12 and the robot 414. In the following description, the data processing device 12 will be referred to as the "server" and the robot 414 will be referred to as the "terminal".

[0148] (Example 1) Since the specific processing flow is the same as that in Example 1 described in the first embodiment above, the description is omitted.

[0149] (Application Example 1) Since the specific processing flow is the same as that in Application Example 1 described in the first embodiment above, the description is omitted.

[0150] The specific processing unit 290 sends the result of the specific processing obtained in Embodiment 1 or Application Example 1 described above to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the controlled object 443 to output the result of the specific processing. The microphone 238 acquires voice representing the user's input on the specific processing result. The control unit 46A sends the voice data representing the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[0151] Furthermore, the processing performed by the data processing system 10 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the robot 414, but it can also be executed jointly by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the robot 414. Additionally, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or external devices, and the robot 414 acquires or collects information required for processing from the data processing device 12 or external devices.

[0152] In the above embodiments, examples of specific processing performed by the data processing device 12 are given, but the technology disclosed herein is not limited to this, and specific processing may also be performed by the robot 414.

[0153] Furthermore, the emotion-specific model 59, serving as an emotion engine, can determine the user's emotion based on a specific mapping. Specifically, the emotion-specific model 59 can determine the user's emotion based on an emotion graph that serves as a specific mapping (see...). Figure 9 The system determines the user's emotions. In addition, the emotion-specific model 59 can also determine the robot's emotions in the same way, and the specific processing unit 290 can also perform specific processing using the robot's emotions.

[0154] Figure 9 This is a diagram representing an emotion map 400 that maps various emotions. In the emotion map 400, emotions are arranged radially from the center in concentric circles. The closer to the center of the concentric circles, the more primitive the emotion is. Further out on the concentric circles, emotions are arranged representing states or actions arising from mood. Emotion is a concept that includes both feelings and mental states. To the left of the concentric circles, emotions generated by reactions occurring in the brain are arranged roughly. To the right of the concentric circles, emotions guided by situational judgments are arranged roughly. Above and below the concentric circles, emotions generated by reactions occurring in the brain and guided by situational judgments are arranged roughly. Furthermore, the emotion of "pleasure" is arranged above the concentric circles, and the emotion of "unpleasantness" is arranged below. Thus, in the emotion map 400, various emotions are mapped according to the structure of emotion generation, while easily generated emotions are mapped nearby.

[0155] These emotions are distributed at the 3 o'clock position on the Emotion Chart 400, and usually fluctuate between peace and unease. In the right half of the Emotion Chart 400, because situational awareness is more dominant than internal feelings, it gives a sense of calm.

[0156] The inner side of the emotion diagram 400 represents the mind, and the outer side of the emotion diagram 400 represents actions. Therefore, the further you go to the outer side of the emotion diagram 400, the more the emotion can be seen (manifested in actions).

[0157] Here, human emotions are based on a balance of various factors such as posture and blood sugar levels. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. In robots, cars, and motorcycles, emotions can also be created based on a balance of factors such as posture and remaining battery power. When these balances deviate from the ideal, it indicates unhappiness; when they approach the ideal, it indicates pleasure. Emotion maps can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Speech Emotion Recognition and Brain Physiological Signal Analysis Systems for Emotion, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). In the left half of the emotion map, emotions belonging to the "Reaction" domain, where sensation is dominant, are arranged. Furthermore, in the right half of the emotion map, emotions belonging to the "Situation" domain, where situational cognition is dominant, are arranged.

[0158] The emotion map defines two types of emotions that promote learning. One is a negative emotion located near the middle of "repentance" or "reflection" on the situation side. That is, when the robot experiences negative emotions such as "I never want to feel this way again" or "I never want to be scolded again." The other is a positive emotion located near "desire" on the response side. That is, when the robot experiences positive feelings such as "wanting more" or "wanting to know more."

[0159] The emotion-specific model 59 feeds user input into a pre-learned neural network to obtain emotion values ​​representing each emotion shown in the emotion graph 400, and determines the user's emotion. This neural network is pre-learned based on multiple learning data sets that combine user input with emotion values ​​representing each emotion shown in the emotion graph 400. Furthermore, this neural network is learned to... Figure 10 As shown in sentiment graph 900, sentiment values ​​in nearby configurations are similar to each other. Figure 10 Examples show that multiple emotions such as "peace of mind", "stability", and "reassurance" have similar emotional values.

[0160] The above description primarily focuses on the functions of the data processing device 12 to illustrate the system disclosed herein. However, the system disclosed herein may not necessarily be implemented on a server. The system disclosed herein can be implemented as a general information processing system. For example, it can be implemented as a software program running on a personal computer, an application running on a smartphone, etc. The methods disclosed herein can be provided to users in the form of SaaS (Software as a Service).

[0161] In the above embodiments, an example of specific processing performed by a single computer 22 is given. However, the technology disclosed herein is not limited to this, and distributed processing of specific processes by multiple computers, including computer 22, is also possible. For example, the data generation model 58 may also be provided in an external device of the data processing apparatus 12, where data is generated based on input data.

[0162] In the above embodiments, an example of storing a specific processing program 56 in memory 32 is illustrated, but the technology disclosed herein is not limited thereto. For example, the specific processing program 56 may also be stored in a portable computer-readable non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 performs specific processing according to the specific processing program 56.

[0163] Alternatively, the specific processing program 56 can be stored in a storage device such as a server connected to the data processing device 12 via a network 54, and the specific processing program 56 can be downloaded and installed into the computer 22 upon request from the data processing device 12.

[0164] Furthermore, it is not necessary to store the entire specific process 56 in a storage device such as a server connected to the data processing device 12 via the network 54, nor is it necessary to store the entire specific process 56 in the memory 32; a portion of the specific process 56 may also be stored.

[0165] As a hardware resource for performing specific processing, various processors can be used. For example, a CPU can be listed as a processor; it is a general-purpose processor that functions as a hardware resource for performing specific processing by executing software, i.e., a program. Furthermore, as a processor, a dedicated circuit can be listed as a processor; it is a processor with a circuit structure specifically designed for performing specific processing, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit).

[0166] Every processor has built-in or connected memory, and every processor performs specific processes by using memory.

[0167] The hardware resources for performing a specific process can consist of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Furthermore, the hardware resources for performing a specific process can also be a single processor.

[0168] As an example of a single processor, the first type consists of a combination of one or more CPUs and software, which functions as a hardware resource to perform specific processing. The second type uses a processor, such as a System-on-a-chip (SoC), which implements the entire system functionality, including multiple hardware resources performing specific processing, using a single IC chip. In this case, the specific processing is implemented using one or more of the aforementioned processors that serve as hardware resources.

[0169] Furthermore, as the hardware architecture of these various processors, more specifically, circuits combining semiconductor elements and other circuit components can be used. Moreover, the specific process described above is merely an example. Therefore, it goes without saying that, without departing from the main point, unnecessary steps can be removed, new steps can be added, or the processing order can be changed.

[0170] The foregoing descriptions and illustrations are detailed explanations of the parts covered by this disclosure and are merely one example of this disclosure. For instance, the descriptions of the above-described structure, function, role, and effect are just one example of the structure, function, role, and effect of the parts covered by this disclosure. Therefore, it goes without saying that, without departing from the spirit of this disclosure, unnecessary parts can be deleted, new elements can be added, or replacements can be made to the foregoing descriptions and illustrations. Furthermore, to avoid confusion and facilitate understanding of the parts covered by this disclosure, explanations of technical common sense that does not require special explanation for implementing this disclosure have been omitted from the foregoing descriptions and illustrations.

[0171] All documents, patent applications and technical standards described in this specification are incorporated herein by reference as if they were specifically and individually described as incorporated by reference.

[0172] In addition, the following notes further explain the above explanation.

[0173] (Note 1) An access management system for data centers includes a user information management department, a permission granting management department, an audit record management department, a generative AI processing department, and an API linkage department, characterized in that... The User Information Management Department uniformly manages detailed attribute information such as the name, ID, role, company, employment status, contact information, shift information, area of ​​responsibility, and past entry and exit history of operation and maintenance technicians, maintenance workers, external contractors, security officers, cleaning staff, and equipment inspectors who enter and exit the data center. The permissions granted to the management department are to use generative AI to comprehensively analyze each user's role, affiliated company, business content, work plan, qualification information, past access performance, etc., and automatically allocate the minimum and optimal logical access permissions and physical access permissions. Furthermore, the API linkage department receives event information such as onboarding, transfer, resignation, and the start and end of work periods in real time, and immediately allocates, changes, or revokes the permissions of the corresponding users. The audit record management department records all user account operations, permission changes, physical room entry and exit records, logical system access records, data synchronization history with external systems via API, terminal information and time information in detail. The generative AI processing unit comprehensively analyzes this information, detects abnormal behavior and incorrect permission settings in real time, and proposes risk notifications and corrective measures to the administrator; and The API linkage unit communicates with external systems such as human resources, access control, equipment management, cloud services, and local systems via API to achieve information synchronization and unified access control across the entire system.

[0174] (Note 2) The system as described in Appendix 1 is characterized in that, The generative AI processing unit analyzes the user's past physical entry and exit history and logical system access history, automatically extracts abnormal patterns such as access in different time periods or areas than usual, external contractors entering suspicious areas, and access attempts by users who have left the company, notifies the administrator in real time, and performs immediate suspension or revocation of the corresponding user's permissions as needed.

[0175] (Note 3) The system as described in Appendix 1 is characterized in that, The permissions granted to the management department use generative AI to learn the user's role, company, work plan, qualification information, past access performance, shift information, and area of ​​responsibility, and dynamically allocate temporary or permanent physical and logical access permissions based on the work period and area of ​​responsibility. When the work period ends or the area of ​​responsibility changes, the corresponding permissions are automatically revoked, and all permission change history is recorded in the audit record management department.

Claims

1. A system comprising a user information management department, a permission granting management department, an audit record management department, and a generative AI processing department, characterized in that, The User Information Management Department uniformly manages the individual attribute information, department, position, and employment status of business commission employees engaged in the communications industry or IT-related businesses; The permission granting management department uses generative AI to automatically formulate the minimum permission granting strategy corresponding to each user's department and position, and links with event information such as onboarding, transfer, and resignation from the human resources system to execute the allocation, modification, and revocation of permissions in real time; The audit record management department records all account operations, permission changes, system access history, and time information together. as well as The generative AI processing unit parses the information from these units, automatically detects signs of incorrect permission settings and unauthorized access, and synchronizes information with the cloud infrastructure and the local environment's authentication infrastructure through API linkage, thereby achieving consistent ID management and reducing security risks in a hybrid environment.

2. The system as described in claim 1, characterized in that, The generative AI processing unit parses the access history and permission change history recorded in the audit record management unit, automatically extracts permission granting and access patterns that are different from the usual ones, and notifies the administrator, thereby realizing the early detection of information leakage and unauthorized access and the automatic proposal of corrective measures.

3. The system as described in claim 1, characterized in that, The permission granting management department uses generative AI to learn from past permission granting performance, business content, and system utilization accumulated by department and position, automatically update the best minimum permission granting strategy, and change or revoke the corresponding user's permissions in real time when receiving transfer or resignation events from the human resources system, and record all such processing in the audit record management department.

Images