Information processing methods, programs, and information processing systems

Abstract

This invention provides an information processing method, program, and system that implement access control while ensuring security and scalability. [Solution] In an information processing system comprising a server as an information processing device, a user terminal, and an agent system, all connected to each other via a network for mutual communication, the information processing method performed by the server includes an information acquisition step of acquiring information about a target entity; a query step of generating query information for estimating a reliability index of the target entity from the information and presenting the query information to the target entity; and an estimation step of acquiring response information from the target entity to the query information and estimating a reliability index of the target entity from the response information.

Description

Technical Field

[0001] The present invention relates to an information processing method, a program, and an information processing system.

Background Art

[0002] Cited Document 1 discloses a network management system in which a communication management server verifies unique identification information of a terminal to be connected, thereby determining whether connection is permitted or not and excluding unauthorized terminals.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of Invention

Problems to be Solved by the Invention

[0004] In the case of the above prior art, since connection permission is determined based on pre-prepared identification information, it is not possible to determine connection permission for a target that has no identification information in the first place, which may result in lack of scalability. There is room for improvement in this respect.

[0005] In view of the above facts, the present invention aims to perform access control that achieves both security assurance and scalability.

Means for Solving the Problems

[0006] According to the information processing method according to one embodiment, the information processing method is executed by an information processing apparatus, and includes: an information acquisition step of acquiring information related to a target entity; an inquiry step of generating inquiry information for estimating a reliability index of the target entity from the information, and presenting the inquiry information to the target entity; and an estimation step of acquiring response information from the target entity in response to the inquiry information, and estimating the reliability index of the target entity from the response information.

[0007] According to a program of one embodiment, the information processing device is made to execute an information processing method which includes an information acquisition step of acquiring information about a target entity, a query step of generating query information for estimating a reliability index of the target entity from the information and presenting the query information to the target entity, and an estimation step of acquiring response information from the target entity to the query information and estimating the reliability index of the target entity from the response information.

[0008] According to an information processing system according to one embodiment, the information processing system executed by the information processing device includes: an information acquisition unit that acquires information about a target entity; an inquiry unit that generates inquiry information for estimating a reliability index of the target entity from the information and presents the inquiry information to the target entity; and an estimation unit that acquires response information from the target entity to the inquiry information and estimates the reliability index of the target entity from the response information. [Effects of the Invention]

[0009] According to one embodiment, access control can be implemented that balances security and scalability. [Brief explanation of the drawing]

[0010] [Figure 1] This figure shows an example of the configuration of an information processing system according to one embodiment. [Figure 2] This figure shows an example of a server hardware configuration according to one embodiment. [Figure 3] This figure shows an example of the functional configuration of a server according to one embodiment. [Figure 4] A flowchart showing an example of a process performed by an information processing system according to one embodiment of this model. [Figure 5] This is a conceptual diagram showing an example of access control according to one embodiment. [Figure 6]This is a conceptual diagram showing an example of access control related to Modification 1. [Figure 7] This is a conceptual diagram showing an example of access control related to Modification 2. [Figure 8] This is a conceptual diagram showing an example of access control related to Modification 3. [Modes for carrying out the invention]

[0011] (Embodiment) An embodiment of the information processing system according to the present invention will be described below with reference to Figures 1 to 5. In each figure, identical or equivalent components and parts are given the same reference numerals. Furthermore, the dimensional ratios in the drawings are exaggerated for illustrative purposes and may differ from actual ratios.

[0012] (System Overview) First, an overview of the information processing system 10 according to this embodiment will be described. The information processing system 10 according to this embodiment exchanges inquiry information with at least one of the target entities, namely the user terminal and the agent system (hereinafter sometimes simply referred to as "agent"), and dynamically estimates the trust index of the target entity by analyzing the content of the response, etc., and provides a mechanism for flexibly controlling access rights and the scope of information disclosure. Specifically, the server first acquires information about the target entity, generates and presents inquiry information based on that information, and further analyzes the response information to derive the trust index of the target entity. A feature of this system is that it enables more granular access control than conventional single-ID matching methods, such as changing whether access is permitted or not according to the obtained trust index. In this specification, "trust index" is a comprehensive numerical or graded evaluation of the trustworthiness and authority level of the target entity (including not only the user terminal and agent system mentioned above, but also users and AI agent systems through the operation of the user terminal), and is an index for objectively showing the legitimacy, authenticity, and degree of authority conformity of the target entity (details will be described later). Furthermore, the term "agent" is not merely a passive terminal, but refers to an entity that actively responds and acts in response to inquiries from a server and access control policies. This also includes agents that utilize pre-trained models (so-called AI agents).

[0013] (System Configuration) Figure 1 shows an example of the configuration of the information processing system 10 according to this embodiment. As shown in Figure 1, the information processing system 10 according to this embodiment comprises a server 12 as an information processing device, a user terminal 14, and an agent 16, all of which are connected to each other via a network N so as to be able to communicate with one another. The network N is, for example, a wired LAN (Local Area Network), a wireless LAN, the Internet, a public telephone network, a mobile data communication network, or a combination thereof.

[0014] The user terminal 14 is an example of an information processing apparatus that is operated by the user U to input and display various types of information. The user terminal 14 may be a PC (Personal Computer), a smartphone, a tablet terminal, a server apparatus, a microcomputer, a wearable device, or a combination thereof.

[0015] The server 12 is an example of an information processing apparatus that acquires information input from the user terminal 14, performs processing based on the information, and outputs a result. The server 12 may be a PC (Personal Computer), a smartphone, a tablet terminal, a server apparatus, a microcomputer, or a combination thereof. Further, the server 12 may be the agent described above. The specific configuration and operation of the server 12 will be described later.

[0016] As described above, the agent 16 is an element disposed separately from the server 12 and the user terminal 14 in the present embodiment, and is an example of a subject that interacts with the system as a target entity. The agent 16 is characterized in that it can be implemented in various specific forms, ranging from a program form of software only, to a physical robot, an IoT device, and even the latest AI agent utilizing a large language model (LLM). The following shows assumed configuration examples, but the present invention is not limited to these examples.

[0017] (1) Software agent Examples include chatbots, script engines, and RPA (Robotic Process Automation) tools that operate over a network. These receive human instructions or task requests from the server 12, access a database as needed, and perform additional calculation and analysis processing. Scale-out is easy, and a configuration in which a large number of instances are operated simultaneously as microservices in a cloud environment can be considered.

[0018] (2) Robot-type agent with hardware As devices with physical entities, this is a case where industrial robots on factory lines, household robots, smart speakers, and the like function as agents. Each robot transmits its own operation logs and sensor information to the server 12, and switches its operation mode according to the response content and access right settings from the server 12, whereby advanced coordinated control and interaction with humans can be realized.

[0019] (3) IoT Device / Edge Agent This is a configuration in which an IoT device incorporating a microprocessor and a sensor and measuring environmental information serves as the role of an agent. These transmit measurement data to the server 12 via the network N, receive control policies and access right settings generated by the server 12, and change their operations. By combining with edge computing, load distribution to the server and improvement of real-time performance can be expected.

[0020] (4) Virtual Agent (Digital Human, etc.) Avatars and digital humans that act in the metaverse or virtual space can also be agents that access the server 12 via speech recognition and natural language analysis. While interacting with the user U in the virtual space, it is possible to dynamically change service contents and display information while receiving inquiries and access control from the server 12.

[0021] (5) Composite / Hybrid Agent Configurations having a hierarchical structure are also assumed, such as linking a software agent and a robot-type agent, or arranging a parent agent that supervises a plurality of IoT devices. For example, at a manufacturing site, a large-scale system can be constructed in which a plurality of autonomous robots each hold their own security authorities, and a higher-level software agent manages a unified access control policy.

[0022] (6) Latest AI Agent Utilizing Large Language Model With the recent advancements in generative AI technology and large-scale language models (LLMs), agents centered around sophisticated natural language processing engines are emerging. These agents dynamically invoke models trained on large datasets in the cloud and respond to requests from users and other agents. Combined with automated integration with external APIs and third-party services, they can autonomously handle complex task flows and possess high adaptability to environmental changes and new information. Compared to traditional rule-based agents or those dependent on limited datasets, they can perform processing with far greater flexibility and richer knowledge and reasoning capabilities.

[0023] As described above, the concept of "Agent 16" can encompass a wide range of entities, from software-centric programs to hardware-based robots, lightweight IoT devices, and even large-scale language model-based AI. In this embodiment, by treating all of these as examples of "target entities," it is possible to flexibly provide query information, access permission settings, and collaboration processes according to the characteristics of each agent. In other words, the term "agent" here is not limited to individual agent structures or specific implementation methods, but includes all forms of agents operating in large-scale and diverse network environments.

[0024] (Hardware configuration) Figure 2 is a block diagram showing the hardware configuration of server 12. Server 12 comprises a processor 120, memory 122, storage 124, communication interface 126, input / output interface 128, and drive device 134, all of which are connected to each other via bus B for communication.

[0025] The processor 120 controls the configuration of the server 12 and realizes the functions of the server 12 by loading various programs stored in the storage 124 into the memory 122 and executing them. The programs executed by the processor 120 include, but are not limited to, the OS (Operating System) and the program 220 described later. By the processor 120 executing these programs, a part of the state visualization method according to this embodiment is realized. The processor 120 is, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), an ASIC (Application Specific Integrated Circuit), a DSP (Digital Signal Processor), or a combination thereof.

[0026] Memory 122 is, for example, ROM (Read Only Memory), RAM (Random Access Memory), or a combination thereof. ROM is, for example, PROM (Programmable ROM), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), or a combination thereof. RAM is, for example, DRAM (Dynamic RAM), SRAM (Static RAM), MRAM (Magnetoresistive RAM), or a combination thereof.

[0027] Storage 124 stores the OS, various programs described later, and various data. Storage 124 is, for example, flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), SCM (Storage Class Memories), or a combination of these.

[0028] The communication interface 126 is an interface for connecting the server 12 to external devices, including user terminals 14 and agents 16, via the network N, and for controlling communication. The communication interface 126 is, for example, an adapter compliant with Bluetooth®, Wi-Fi®, ZigBee®, Ethernet®, or optical communication (e.g., Fibre Channel).

[0029] The input / output interface 128 is an interface for connecting input devices 132 and output devices 130 to the server 12. Input devices 132 include, for example, a mouse, keyboard, touch panel, microphone, scanner, camera, various sensors, operation buttons, or a combination thereof. Output devices 130 include, for example, a display, projector, printer, speaker, vibrator, or a combination thereof.

[0030] The drive device 134 reads and writes data to the disk media 136. The drive device 134 is, for example, a magnetic disk drive, an optical disk drive, a magneto-optical disk drive, or a combination thereof. The disk media 136 is, for example, a CD (Compact Disc), a DVD (Digital Versatile Disc), an FD (Floppy Disk), an MO (Magneto-Optical disk), a BD (Blu-ray® Disc), or a combination thereof.

[0031] In this embodiment, the program may be written to memory 122 or storage 124 during the manufacturing stage of server 12, or it may be provided to server 12 via network N, or it may be provided to server 12 via a non-temporary, computer-readable recording medium such as disk media 136.

[0032] Furthermore, since the hardware configuration of the user terminal 14 is substantially the same as that of the server 12 described above, a detailed explanation will be omitted.

[0033] (Functional Configuration) Next, the functional configuration of server 12 will be described. Figure 3 shows an example of the functional configuration of server 12. When executing various programs, server 12 uses the above hardware resources to realize various functions. Server 12 has a storage unit 22, a control unit 24, and a communication unit 20 as the functional configuration that server 12 realizes. Each functional configuration is realized when the processor 120 reads and executes a program 220 stored in memory 122 or storage 124.

[0034] The communication unit 20 is implemented by the communication interface 126. The communication unit 20 sends and receives information with the user terminal 14 via the network N. The communication unit 20 receives information input from the user terminal 14. The communication unit 20 also sends information to the user terminal 14 and receives requests from user U from the user terminal 14.

[0035] The memory unit 22 is implemented by memory 122 and storage 124. The memory unit 22 stores the program 220, the trained model 222, the target entity information DB 224, the query information DB 226, the response information DB 228, the trust index DB 230, the access control DB 232, and the user information DB 234.

[0036] The program 220 is read and executed by the processor 120 of the server 12, and is responsible for realizing various procedures of the information processing method according to this embodiment.

[0037] The pre-trained model 222 is configured to be used in this system to analyze the content of natural language queries and responses in a more advanced and flexible manner. In this system, the response patterns and usage history of the target entity are prepared as initial training data, and a loss function is calculated that shows the error between the response content and the correct label or expected response level, and the parameters are updated. During training, the aim is to improve the accuracy of contextual understanding and response estimation that goes beyond simple keyword matching by combining natural language expressions that are difficult to distinguish and corpora that contain diverse linguistic features. For example, in order for the pre-trained model 222 to be able to infer with a certain degree of confidence how reliable the ambiguity of the intent presented by the target entity is, elements specific to the actual usage environment such as redundant expressions, mistranslations, and omissions are incorporated into the training data, and corrections are made step by step while observing the difference from the correct data. In the training phase, it is also envisioned that large-scale training may be performed by integrating server 12 logs, past matching results, and existing general-purpose natural language data, and in that case, the error with samples that have been given correct labels is repeatedly calculated to optimize parameters such as weights and biases in the model.

[0038] On the other hand, during the inference phase of the trained model 222, text, conversation history, or a summary of the execution results obtained from the target entity can be incorporated, and reliability can be immediately estimated or the validity of permission settings can be evaluated using the feature representations constructed during training. At this time, the model takes into account the ambiguity and polysemy inherent in natural language, and processes the estimation of the consistency of the response information by considering differences and context that cannot be captured by simple keyword matching. Furthermore, even when using a large-scale language model (LLM), it is possible to perform inference only by reusing the trained parameters without performing fine tuning on a module-by-module basis.

[0039] In other words, the trained model 222 supports two processing steps: query information generation and response information analysis. In the query information generation stage, it refers to responses and historical data previously obtained from the target entity (in this embodiment, agent 16, hereinafter abbreviated) to determine what additional questions should be asked to more clearly confirm the truthfulness of the target entity and the validity of its authority. For example, if a previous response states that it "has level 2 authority" but provides few actual usage examples or supporting evidence, the trained model 222 can use cumulative natural language analysis to generate questions that are likely to reveal contradictions, such as "requesting examples of actually exercising level 2 authority."

[0040] Here, as a concrete example of query information, we can cite the case of presenting a question expressed in natural language. For example, questions such as "What exactly does the Level 2 permission you hold refer to?" or "Please describe the specific process in which you exercised that permission" allow the user to confirm the content of the permission claimed by the target entity in natural language. In this way, generating query information that includes natural language makes it easier to obtain information that cannot be captured by simple numerical or flag matching.

[0041] Next, it is conceivable that the system would incorporate historical data, including past operations performed by the target entity and past dialogue logs, to determine the query information. For example, if a previous response stated that the entity had "never accessed confidential data," but the current response states that it "frequently views confidential folders," the trained model 222 would construct a question that takes these historical inconsistencies into account and ask which information is correct. Specific examples include questions such as, "In our previous interaction, you said you hadn't viewed confidential folders, but now it appears you are viewing them frequently. What circumstances have caused this change?" or, "In your previous response, you said you had 'no history of viewing confidential data,' but recent logs show several instances of viewing. Could you explain?" By prompting the user to compare past statements with the latest responses, a mechanism can be introduced to verify the consistency of the information presented by the target entity.

[0042] Furthermore, when generating query information, it is crucial to create questions that anticipate the correspondence with the response information. To map whether the current response is consistent with the answers obtained from the previous query, or whether further verification is needed, the trained model 222 can incorporate a mechanism to input new query sentences based on the perspective of "how significant should we evaluate a discrepancy between the current response and the previous response?". This is expected to improve the estimation accuracy of subsequent steps, as the confidence index will increase if consistency is confirmed and decrease if a discrepancy is found.

[0043] Furthermore, query information is not just a simple "question," but sometimes includes prompts for actions to be performed on the target entity. For example, it might request a specific action, such as "On a system for which you have permissions, view the configuration file and return its hash value," with the result (hash value, log information, etc.) as the response information. In this case, whether or not a processing result was obtained is considered the answer to the query information, and the trained model 222 can analyze the returned artifacts to confirm whether there is "evidence that the permissions were actually used."

[0044] Furthermore, to verify the reliability of authorization, the trained model 222 sometimes generates queries that confirm whether the target entity has actually had experience accessing a given system. For example, if an entity answers that it has logged into the accounting management system in the past, but that login history cannot be found in the system audit logs, a leading query such as "Do you have any evidence files or records of performing any operations on the accounting management system?" can be presented. This encourages the target entity to provide specific evidence of the authorizations it claims to possess, making it easier to determine their veracity.

[0045] The trained model 222 can also formulate additional questions tailored to the role and scope of work claimed by the target entity. For example, it can automatically generate questions such as, "You said you belong to the security team and can manage cryptographic keys, but specifically which keys have you rotated in the past?" to understand the target entity's history of key management. If the security audit team does not have logs related to the key in question, it can serve as a starting point for determining that the answer is likely false.

[0046] Furthermore, instead of simply asking about the content of the response, the level of proficiency in using the system can be evaluated and confirmed in the context of actual operation and operational situations. For example, instead of asking a general knowledge of permissions, the user should answer with the actual operations and names they need, such as, "How many types of API calls are available at permission level 2? Please give specific endpoint names." If there are inconsistencies or errors in the response from the target entity, the trained model 222 will use the consistency as an evaluation criterion to calculate a confidence index, making it easier to detect potential permission fraud.

[0047] If the target entity claims to have used multiple systems, questions such as "Specifically, which system did you work on, and for what period?" and "What are the details of the projects you handled at that time?" are generated, and the answers are compared with project management information and user activity history from previous logs. For example, if the target entity answers that they "worked on Project X for over a year," but the project management information shows that the user has never been registered, the system will detect the inconsistency and work to lower the confidence metric. Conversely, if the entity provides extremely accurate project names and period information, the trained model 222 can determine that "the answer matches the actual results" and provide a trigger to increase the confidence metric.

[0048] On the other hand, when calculating the confidence index after obtaining the response information, for example, the trained model 222 compares the new response with past responses to detect any inconsistencies or contradictions. If the model denies permissions that were previously granted, or fails to reconcile a case that clearly contradicts a previous claim, a mechanism can be built to lower the confidence index according to the degree of the inconsistency. In short, by incorporating questions that consider "consistency with past responses" at the query information generation stage, the model actively identifies potential inconsistencies and missing information that may arise during the response, making it easier for the trained model 222 to evaluate the degree of inconsistency and contextual inconsistencies in the next step. By supporting this dynamic question-and-answer process, the trained model 222 can accurately determine the accuracy of the target entity and the validity of the permission level, and as a result, the confidence index reflected in the access control DB 232 can be calculated in a more reliable manner.

[0049] For example, if a target entity responds to a query expressed in natural language (see Claim 2) by stating, "I was browsing a confidential folder using Level 2 privileges earlier," but then later says in the same dialogue session, "Actually, I've never browsed it," the trained model 222 detects the discrepancy between the most recent response and past logs. If the degree of inconsistency is high, the model performs a score reduction by lowering the confidence index while comparing it with the response information DB 228 (which includes past response history and dialogue logs, see Claim 3). Ultimately, it is possible to trigger a retry of the query information, such as by adding supplementary questions to determine whether the inconsistency is intentional false reporting or a simple memory lapse.

[0050] Furthermore, the trained model 222 can also evaluate whether the response adequately covers the core of the question by comparing it with the relationship between the question and the response information that was embedded when the query information was generated (see claim 4). For example, if the query is "Please show me the log of the specific operation you performed using Level 2 privileges," and the answer given is an ambiguous "I don't know what the operation was," the model will find from the dialogue history that "in a previous response, the user made statements that suggested they knew the specific operation," and will treat this as an inconsistency. This allows for an evaluation of the completeness and consistency of the response from the target entity, and can be used as a basis for raising or lowering the confidence index.

[0051] Furthermore, if a query is made in a "process prompting format" (see claim 5) based on past responses and operation history, and response information is returned as a result, the trained model 222 verifies the validity of the processing content. For example, it might issue an instruction such as, "If you truly have level 2 privileges, view the specified file and return the hash value," and compare the hash value stored in the response information DB with various logs and file management DB on the server 12 side. If a match is confirmed, it works to raise the confidence index; if there is a mismatch or refusal to respond, it is considered that the user may be overstating their privileges, and a process is applied to lower the confidence index.

[0052] Furthermore, the trained model 222 updates the trust index for the target entity, which is shown as a numerical value or level classification, based on the results of a comprehensive evaluation of the inconsistencies and contradictions described above. This judgment is ultimately reflected in the access control database, allowing for flexible operation, such as raising the permission level of the target entity itself or temporarily strengthening restrictions as a warning. For example, if an entity repeatedly provides consistent answers to many questions, its trust index will cumulatively increase, granting it permission to access highly sensitive data. Conversely, if there are a series of contradictory answers in a short period, the system can forcibly set the permission level to a lower level to ensure system security.

[0053] In addition, the trained model 222 performs more flexible estimations by referring to subtle differences in expression found in the target entity's response sentences and changes in responses over time. For example, it is possible to evaluate whether a response is fundamentally different from past statements, taking into account the ambiguities inherent in natural language, such as "whether the meaning is inconsistent despite the use of synonyms" and "whether there are substantial contradictions due to abbreviations or changes in technical terms." Such analysis dealing with subtle differences in wording and context can be performed using the natural language processing capabilities of the trained model 222. This configuration makes it possible to estimate a confidence index for target entities for which there is no past response content. In this embodiment, the trained model 222 is stored in the memory unit 22, but it is not limited to this configuration, and it may also be stored on an external server and used from there.

[0054] The target entity information DB224 stores basic information and past usage history about the target entity. Server 12 can refer to the target entity information DB224 to understand the background of the target entity and select query information and access control policies as needed.

[0055] The query information DB226 stores not only templates and standard phrases used when generating queries, but also logs of queries generated and presented in the past, as well as information on their variations. Specifically, it stores newly created question sentences as the interaction with the target entity progresses, and historical data for reusing content similar to previous queries. Server 12 not only extracts appropriate text from here, but also refers to what questions the target entity has asked and what answers it has given in the past, enabling it to dynamically create queries that are tailored to the dialogue context and the entity's specific history, rather than simply reusing templates. This functionality is useful in situations where it is possible to generate questions that are more likely to detect and induce duplication or inconsistencies with past interactions, or to efficiently repeat additional queries to delve deeper into answers.

[0056] The response information DB228 stores all responses and operation results obtained from the target entity. The stored data is expected to be in various formats, such as natural language text and system data, and is also used when the server 12 refers to the response information to calculate a confidence index.

[0057] The Trust Index DB230 is a database that stores trust indexes and calculation history for each target entity. Server 12 stores the trust index calculated based on response information and other factors in this DB, which can be used to dynamically determine access permissions and authorizations.

[0058] The access control DB232 is a database that stores information defining access scopes and privilege levels according to trust indicators, along with organized history. Server 12 refers to the access control DB232 and appropriately switches the scope of access that target entities can access, maintaining a balance between the overall system load and security.

[0059] User information DB234 is a database that stores user account and attribute information. When the target entity behaves as a user, detailed individual settings and permission management can be performed based on the information stored here, allowing for more granular access control on server 12.

[0060] The control unit 24 is realized when the processor 120 reads and executes the program 220 from the memory 122 (see Figure 2) and cooperates with other hardware components. In this embodiment, the control unit 24 includes an information acquisition unit 242 that collects information about the target entity, an inquiry unit 244 that generates query content based on the acquired information, an estimation unit 246 that analyzes response information and calculates reliability indicators, and an access control unit 248 that sets and updates access rights according to the calculation results. The functions of each part will be described below.

[0061] The information acquisition unit 242 executes the "information acquisition step" described in claim 1. That is, it is the part that acquires information about the target entity. It includes an interface that enables data linkage with various information sources such as external systems and user input, and the acquired attribute information and history information are used for processing by the subsequent inquiry unit 244 and estimation unit 246. In this embodiment, the information acquisition unit 242 may be configured to automatically acquire the target entity's profile, past access logs, etc., by providing functions for external API calls and data reception via the network.

[0062] The inquiry unit 244 executes the "inquiry step" described in claim 1. That is, it generates specific inquiry content by referring to the attribute information and past history of the target entity obtained from the information acquisition unit 242. In this embodiment, natural language processing technology using the trained model 222 is utilized to automatically assemble inquiry information that includes information expressed in natural language by combining it with templates and boilerplate text. The inquiry unit 244 also handles the selection logic of templates stored in the inquiry information DB 226 and controls such as eliminating duplicates with previously used inquiry information, thereby presenting more accurate inquiry information to the target entity and enabling highly accurate estimation of reliability indicators. Note that the inquiry information may be newly generated without combining templates or boilerplate text. Furthermore, the inquiry information in this embodiment may not only include information expressed in natural language, but also include machine-readable information such as numbers, codes, and JSON format, and generate inquiry information that requests a response in natural language. For example, a query may be made that requests a detailed description in natural language of the reason why the permission is needed and the assumed risks, along with a code automatically assigned to identify a specific access permission. This configuration allows for supplementing contextual information that cannot be obtained through conventional, standardized data input alone, enabling a more accurate estimation of the reliability metrics of the target entity.

[0063] The estimation unit 246 is responsible for executing the "estimation step" defined in this embodiment. That is, it performs a series of processes to obtain response information from the target entity to the query information, analyze its contents, and calculate a confidence index. Upon receiving the response to the query information generated by the query unit 244, it formats the response data through natural language processing and format conversion, and then calls the trained model 222 to perform inference processing. Specifically, it calculates a confidence index for the target entity by quantifying or hierarchically evaluating at least one of the following: inconsistencies in the content of the response information itself, consistency between the response information and past history information, and attribute information obtained from external sources. Furthermore, a more precise evaluation may be performed by weighting and scoring multifaceted elements such as the basis for the response content and the presence or absence of third-party corroboration. The operation of the estimation unit 246 will be explained below with more specific examples.

[0064] For example, when checking the consistency between the response information submitted by a target entity and its past history, the system can automatically determine whether the response contradicts previously provided information by using an analysis logic that matches inconsistencies in keywords and date information within the text. If the response matches a past response and the newly added information also supports the past response without contradiction, a positive score is awarded. For example, if a response is provided that reaffirms the correctness of previously provided authentication information, and details such as the issuing authority and acquisition date of that authentication information perfectly match those of the past, a higher level of consistency is recognized, and the score is increased to raise the confidence index. On the other hand, if the response differs significantly or a significant discrepancy is found, a negative score is added according to the amount and importance of the difference. This step-by-step control ensures that minor variations in expression do not result in large deductions, while reliably lowering the score when there is a substantial contradiction.

[0065] Furthermore, an example is given of comparing the attribute information of the target entity (such as organization name, department name, and area of ​​activity) with external public databases and authentication services to evaluate whether there are any inconsistencies according to the permission level. For example, the organization's qualification list or employee roster can be obtained as external information, and it can be checked whether the responded qualification name is actually a valid and registered one. Based on this result, the confidence index of the target entity can be adjusted in the positive or negative direction to derive the final confidence index. A large positive score is assigned when the validity of the qualification can be clearly verified, and a large negative score is added if there are deficiencies such as unregistered status or roster mismatch.

[0066] Furthermore, if the response information includes multiple documents or elements, natural language processing is applied step-by-step to classify and summarize the semantic structure of the text in detail, and then compared with various patterns learned by the trained model 222. For example, the frequency of use of honorifics and technical terms in the other party's utterances, and the logical consistency of the response are analyzed, and these are quantified and combined to calculate a more accurate confidence index. By assigning positive scores to responses that are logically consistent and negative scores to ambiguous or contradictory parts, the reliability of the response can be grasped from multiple perspectives.

[0067] The aforementioned reliability indicators are also used for access control and conditions for subsequent queries, playing a role in continuously and dynamically optimizing interactions with the target entity. In this embodiment, the estimation unit 246 integrates these multifaceted analysis methods to meticulously calculate the reliability indicators of the target entity, enabling smooth control of the subsequent processing flow. For example, if it is determined that the logic and basis of the response are sufficiently consistent (the reliability indicator is positive), the frequency of additional queries may be reduced. On the other hand, if there are many inconsistencies (the reliability indicator is negative), additional queries can be continuously made to scrutinize the situation. This allows for dynamic flow management based on the fluctuations in the reliability indicators.

[0068] The access control unit 248 sets access privileges based on the trust index calculated by the estimation unit 246. In this embodiment, the privilege level of the target entity is updated in real time, and access to various resources within the system is dynamically assigned. Furthermore, the access control unit 248 can automatically perform processes to grant higher privileges, such as raising the privilege level (expanding the scope of access and operable functions), in conjunction with the recording of usage status and estimation results, when certain conditions are met. On the other hand, it is also possible to reduce or revoke privileges when the trust index is below a certain condition. As an example, as shown in Figure 5, the server 12 estimates the trust index based on the response information from target entities A, B, and C (in the figure, target entity A has a trust index of L2, target entity B has a trust index of L1, and target entity C has a trust index of L0). Furthermore, the databases that server 12 can access are pre-defined as highly confidential data, semi-confidential data, and general data. The access permissions for each are set as follows: trust index L3 or higher for highly confidential data, trust index L2 or higher for semi-confidential data, and trust index L1 or higher for general data. The access control unit 248 controls whether or not to allow access to the aforementioned databases according to the trust index of each target entity.

[0069] (Processing performed by the information processing system 10) Figure 4 is a flowchart showing an example of the processing flow in this embodiment. In this system, the processor 120 loads the program 220 into memory 122 and executes it, performing the processing shown below. Although not shown in the figures, if the processor 120 receives information about the termination of operation of the information processing system 10, or information about the termination of operation from the user terminal 14 during the execution of a determination process (these are simply referred to as "termination operations"), it terminates the processing based on the program 220 that is currently being processed.

[0070] The processor 120 acquires target entity information (step S100). Step S100 corresponds to the "information acquisition step" described in claim 1, and acquires attribute information and past history as information about the external system (agent) and user terminal 14 as the target entity. For example, the processor 120 accesses a predetermined address using HTTP or other communication protocols and sends a request including the user ID to collect attribute information of the target entity. The received response includes the role of the target entity and the identifier of past logs, etc. The processor 120 converts and parses this, and then records (1) the request transmission time, (2) the response reception time, (3) the status, etc. in the "information acquisition log" table to indicate the completion of this step. Next, the processor 120 generates query information (step S102). Based on the acquired information about the target entity, it creates query information using natural language or standard phrases. For example, it assigns a query ID that uniquely corresponds to the query information, and adds a new record that associates the query ID with the target entity ID to the "query response log," making it easier to track responses in subsequent steps.

[0071] The processor 120 presents the query information (step S104). That is, it presents the previously generated result to the target entity by means of transmission or screen display. Steps S102 and S104 correspond to the "query step" described in claim 1. For example, the processor 120 sends an HTTP POST request to a predetermined address, and the transmitted content includes the query ID, the question text, or the payload of the task request. Alternatively, in the case of a user terminal 14, the query text and query ID may be displayed on the screen, and a UI may be deployed that allows the user to input an answer. After that, the processor 120 determines whether or not there is response information (step S106). If there is response information (step S106: YES), the processor 120 proceeds to step S108, which will be described later. On the other hand, if there is no response information (step S106: NO), the processor 120 moves to step S104. As an example, although not shown in this embodiment, if no further response is received, the system is configured to record "ERR_TIMEOUT" in the log and display a message on the user terminal 14 indicating that the inquiry ID has timed out.

[0072] The processor 120 acquires response information (step S108). That is, it receives the response sent from the target entity and performs format conversion and pre-analysis formatting as necessary. For example, upon receipt, it stores items such as "response_id" and "answer_data" included in the response, linked to the query response log. At the same time, it records the date and time the response was received and the response status in the log, and manages them in a way that is easily observable from the outside. Next, the processor 120 estimates a confidence index (step S110). That is, it calculates the authority level and trustworthiness of the target entity based on the consistency of the acquired response information itself and its consistency with past history information. For example, it matches the ID included in the response time and response content with the operation log and attribute information of the target entity acquired in step S100, and analyzes the consistency score using a trained model to output a confidence index as a numerical value (0 to 100) or level (L0 to L3). In this embodiment, the finally calculated confidence index is written to the query response log, etc., and notified via a predetermined address as a screen display or API response. Steps S108 and S110 correspond to the "estimation step" described in claim 1.

[0073] The processor 120 sets access privileges (step S112). That is, it dynamically assigns the scope of access permissions and operational privileges according to the previously estimated trust index. For example, if the trust index is high, it grants advanced privileges, and if it is low, it restricts access to only general privileges, updating the access control DB in real time. Note that step S112 corresponds to the "privilege setting step" described in claim 6. Next, the processor 120 further performs access control (step S114). That is, it performs actual resource utilization and operation restrictions based on the set privilege level. For example, the server 12 queries the access control DB and monitors the request of the target entity. If the request is within the permitted range, it returns an ACK response (ACK is an abbreviation for "Acknowledgment," which refers to "a response to confirm that transmitted information has been received and processed" in communication protocols and various systems). If it is not permitted, it returns an error message or a re-authentication request via a predetermined address. This makes it easy for external audits to understand when and at what level each entity was able to access resources by referring to the access control history and logs.

[0074] Once the processor 120 has completed the access control execution in step S114, it terminates the series of processes based on the program.

[0075] (Effects of the embodiment) According to the information processing system 10 of this embodiment, the server 12, acting as an information processing device, performs the following steps: an information acquisition step to acquire information about a target entity; an inquiry step to generate inquiry information for estimating the reliability index of the target entity from the information and present the inquiry information to the target entity; and an estimation step to acquire response information from the target entity to the inquiry information and estimate the reliability index of the target entity from the response information. In other words, it works to flexibly and efficiently grasp the reliability index of the target entity through the exchange of inquiry information. Specifically, by making some kind of inquiry or task request to the target entity as inquiry information and comprehensively evaluating the response content, response speed, accuracy and consistency of the response, it becomes possible to make detailed judgments to increase or decrease the potential risk and trustworthiness of the target entity. Furthermore, a flexible verification process can be implemented according to operational requirements and the characteristics of the target entity. Thus, a key feature is that it can dynamically and with high scalability perform permission verification, which previously relied on manual or fixed judgment logic. As a result, adaptive access control (which balances security assurance and scalability) can be implemented for the target entity without compromising scalability and ensuring security. Here, "adaptive" refers to the ability to flexibly adapt and respond to changes in circumstances and the environment. Furthermore, "ensuring security" refers to taking necessary technical means and operational measures to maintain a state in which information and systems are appropriately protected from threats such as unauthorized access and tampering.In addition, although this embodiment mainly describes an example in which the server 12 performs the information acquisition step, query step and estimation step as described above, it is not limited to this, and other information processing devices, such as the user terminal 14, may share and perform some or all of these steps.For example, a configuration in which steps are shared among multiple devices via a network is also conceivable, such as the user terminal 14 handling part of the information acquisition step and the server 12 performing the estimation step.

[0076] The effects described above can also be rephrased as follows: (1) Visualization of the behavior of the target entity By repeatedly generating and presenting inquiries and analyzing the responses, it becomes possible to understand the behavior and characteristics of target entities (humans, AI agents, etc.) from a different perspective than static profile information. For example, metrics such as "variability in response time," "consistency of response content," and "consistency with past data" can be used to more clearly visualize the actual behavior patterns of entities. (2) Building a foundation for dynamic evaluation Instead of a one-time information input, the system incorporates interactive inquiry steps, allowing for a gradual scrutiny of the entity's trustworthiness by tracking multiple responses. This mechanism facilitates long-term, intermittent observation, making it easier to capture changes in factors such as risk level and proficiency in real time and in stages. (3) Early detection of suspicious responses By analyzing the accuracy and inconsistencies of responses, suspicious trends that cannot be captured by simple static ID matching can be identified early on. Specifically, if clear errors in the content of the response, misuse of technical terms, or significantly delayed responses are repeated, the confidence index can be lowered each time, ultimately leading to a decision to prompt a different procedure (however, for the purposes of this explanation, this decision is considered separately from "using estimation results"). (4) Incorporation of diverse evaluation elements Responding to inquiries can easily integrate not only natural language information but also, in some cases, behavioral logs and additional attributes (such as biometric authentication and sensor information). This is because there is a foundation for scoring multiple elements in the form of trust metrics, making it easy to add additional evaluation criteria later. This facilitates future feature expansion and integration with other authentication methods. (5) Improvement of operational efficiency Establishing a system that automatically and progressively generates queries and analyzes responses will significantly reduce manual checks by operators while still allowing for the necessary accuracy assessments. This will minimize the need for manual verification, leading to reduced human resources and lower overall operating costs.

[0077] Thus, the very fact that this embodiment "estimates trust indicators through interaction with the target entity" brings advantages such as the incorporation of various additional elements like biometric authentication and behavioral logs, and a departure from fixed evaluation models. Ultimately, the effectiveness at this stage lies in enabling multifaceted and continuous state understanding. In other words, once this estimation mechanism is in place, a foundation is created that makes it easier to add further extensions (such as access control) later on. One of the advantages is that it allows for accurate evaluation of the entity's state in a way that goes beyond mere static information verification.

[0078] Furthermore, according to the information processing system 10 of this embodiment, the query information includes information expressed in natural language. In other words, because queries and responses are performed using natural language, it has the effect of obtaining detailed information that takes advantage of the expressiveness of the target entity. By communicating in natural language, it is possible to grasp situations and nuances that cannot be captured by a checklist of fixed items alone, and to more deeply examine ambiguities and inconsistencies in the response content. Moreover, as mentioned above, by dynamically switching the format of the query information and generating additional questions while referring to past interactions, it is possible to evaluate the attitude, knowledge level, and potential risk tendencies of the target entity from multiple perspectives. This makes it possible to perform more adaptive access control that ensures security without compromising scalability.

[0079] Furthermore, information about the target entity includes a history of the target entity's output information. This configuration allows for the generation or analysis of query information while referring to records of past interactions. This enables more adaptive access control that ensures security without compromising scalability.

[0080] Furthermore, in the estimation step, the reliability index of the target entity is estimated based on the relationship between the inquiry information and the response information. For example, by comparing the intent and expected response framework of the presented inquiry information with the content included in the actual response in detail, and analyzing it including contextual consistency, the completeness of the information, and the time factor required for the response, it is possible to evaluate the target entity's level of understanding, sincerity, and the possibility of intentional information concealment from multiple perspectives. In this process, the reliability index is increased when the response is highly consistent and specific, while it is decreased when the response is far removed from the intent of the inquiry or is insufficient. In addition, by analyzing the logical structure of the entire response and quantifying its relationship with the surrounding context, and calculating multiple indicators in a comprehensive manner, it becomes easier to infer not only the superficial correctness or incorrectness, but also the motivations and background information hidden behind the response. This enables precise estimation that delves deeply into the relationship between inquiry information and response information, and allows for access control that more effectively balances security assurance and scalability.

[0081] Furthermore, the inquiry information is information that prompts the target entity to perform a task, and the response information is information about the processing results of the target entity. This configuration allows for the estimation of reliability metrics while directly confirming the actual operations performed by the target entity and the completion status of tasks. For example, it becomes possible to evaluate how the target entity executes instructions on the display screen, or whether it can execute them appropriately according to defined specifications and procedures, from response logs and generated deliverables. If the processing procedure is executed accurately and quickly, a high reliability metric is assigned, and conversely, if the procedure is not followed thoroughly or the results are inconsistent, the reliability metric is set lower, enabling empirical and dynamic evaluation. In addition, by accumulating the execution status of the process as a history and analyzing it together with past task execution results and improvement trends, it is possible to compare and verify the information presented by the target entity with its actual actions, contributing to the calculation of a more objective reliability level. As a result, it is possible to perform access control that more effectively balances security assurance and scalability by directly verifying the execution ability, compliance awareness, and work quality of the target entity, rather than merely checking the consistency of static responses.

[0082] Furthermore, this embodiment includes a permission setting step that sets access rights to the target entity according to a trust index. This configuration dynamically adjusts access rights based on the calculated trust index. For example, when the target entity obtains a high trust index, the information it can handle and the functions it can operate can be expanded, while when the trust index declines, the scope of access is immediately restricted, automatically implementing flexible control according to the situation and evaluation. In addition, by also referring to past permission update history and the target entity's processing performance, the system can comprehensively judge the maturity and risk level of the target entity and then subdivide and switch permission levels in stages. This real-time and highly controlled permission setting reduces the risk of potential fraud and information leakage while maintaining the efficiency of information use and work processes within the organization. This enables access control that achieves an even higher level of balance between security assurance and scalability.

[0083] In the above embodiment, query information is generated to include information expressed in natural language. However, the system is not limited to this; query information may also be presented in the form of input selected from predefined options, semi-structured data including numbers, codes, and tags, or structured data such as tables and relational databases. For example, combining boilerplate text and multiple-choice queries can automate and speed up the processing flow, increasing the availability of responses. Generating query information in structured or semi-structured data makes it easier to automatically analyze response content and streamlines integration with external systems. Furthermore, presenting query information in a machine-readable format allows for easy integration with computer-to-computer communication protocols, facilitating data linkage with external systems and the use of analysis tools. In addition, by introducing these variations, multiple formats other than natural language can be used depending on the situation, enabling flexible access control tailored to the diverse characteristics and constraints of the target entities. As a result, this system can further enhance the versatility of queries while still maintaining precise reliability indicator calculations and secure access management. Semi-structured data refers to a format where the entire dataset does not strictly conform to a table format, but it partially indicates hierarchical relationships and connections between elements by including metadata such as tags and keys (for example, XML and JSON).

[0084] Furthermore, the response information to the inquiry information may not only be information expressed in natural language, but may also be so-called multimodal information, which may be at least one of audio, images, and video, or a combination thereof. For example, configurations that handle only audio, configurations that analyze only images, and configurations that process multiple media information simultaneously, such as audio and images, or images and video, can be flexibly selected according to the usage. For example, by using a system that has an audio recognition module on the server to convert audio input into text and a vision analysis module to extract facial and movement features from images and videos, it is possible to grasp the state and reactions of the target entity more comprehensively than information obtained by either means alone. As a result, even if a positive response is given on the surface, if obvious tension or unnatural movements are detected in the video, it becomes easier to judge it as fraud or risk and insert additional verification steps, making it easier to implement advanced security controls.Therefore, this configuration, which includes cases using at least one of audio, images, and video, or cases using multiple of them simultaneously, allows for the rapid execution of appropriate access control and permission settings while evaluating the situation of the target entity from multiple angles.

[0085] Furthermore, in the above-described embodiment, the information regarding the target entity includes the history of the target entity's output information. However, it is not limited to this; it does not have to include the history of the target entity's output information, nor does it have to include other information. Examples of other information include referencing external audit logs, access records, or historical information managed by third-party services. For example, past authentication procedures and transaction information can be obtained via API and centrally integrated as part of the historical data. It is also possible to employ an implementation that cross-references multiple information sources, including related user groups and operation logs of related systems other than the target entity. By adopting such a wide range of variations, it becomes possible to grasp the actual state of the target entity in more detail and accurately, and to provide multi-layered support for estimated reliability indicators. Moreover, the method for accumulating information including such history can utilize not only structured / semi-structured formats such as databases and file storage, but also highly flexible platforms such as object storage in cloud environments and distributed file systems. In this way, by incorporating methods that include not only historical information linked to the target entity, but also various log records and external verification results, the security assurance and scalability of the entire system can be further improved. In addition, the configuration may also include information about the target entity, such as its country or region of origin, the characteristics of the device it uses, and metadata acquired in the past (browser type, OS version, destination IP address, etc.). Furthermore, by combining this with publicly available information such as social media posts and blog articles of the target entity, and operation logs and error logs from internal corporate systems, compliance risks and the presence or absence of fraudulent activity can be grasped in more detail. Moreover, it is possible to introduce a means to comprehensively manage information linked to the target entity by cross-referencing digital data obtained by scanning business transaction records and contract documents. In this way, by utilizing diverse data obtained from multiple sources, it becomes possible to evaluate the characteristics of the target entity from multiple perspectives and accurately, and to further improve the accuracy of the final reliability indicator estimation.

[0086] Furthermore, while the above-described embodiment is configured to estimate the trust index based on the context of the conversation with the target entity (the relationship between inquiry information and response information), it is not limited to this, and the trust index may be estimated using other information. For example, one method is to use non-conversational information such as the target entity's behavior logs and authentication history to perform evaluation (estimation of the trust index). For instance, evaluation criteria could be set such as aggregating the number of login failures and the frequency of operational errors, and lowering the trust index if these values ​​exceed a certain level. In addition, an approach could be considered that links scores obtained from conventional credit information and transaction history, and calculates the trust index comprehensively by referring to credit check models of financial institutions and companies. Moreover, the strength of the authentication level of the device being used (such as biometric authentication or terminal-specific encryption keys) can be monitored, and the trust index can be set higher if access is made from a device with advanced authentication functions. Conversely, for access from a public environment or a terminal whose security has not been verified, the trust index can be suppressed by assigning a temporary risk flag. Furthermore, the target entity's position and authority history within the organization can be checked as appropriate to confirm whether any serious compliance violations have occurred within a certain period. By combining this dynamic and static information, it becomes possible to evaluate the behavior and environmental factors of the target entity from multiple perspectives, even in a way that is independent of the context of the conversation. Incorporating these various additional methods makes it easier to generate appropriate trust indicators not only from the perspective of the consistency of the dialogue content, but also from the perspectives of security and business risk management, thereby achieving even more accurate access control.

[0087] Furthermore, in the above-described embodiment, the query information includes information prompting the target entity to take action, and the response information includes information regarding the processing result of the target entity. However, the query information does not have to include information prompting the target entity to take action. Similarly, the response information does not have to include information regarding the processing result of the target entity.

[0088] Furthermore, while the above-described embodiment performs a permission setting step to set access rights for the target entity according to the trust index, it is not limited to this, and a variety of control policies beyond granting and revoking access rights may be implemented. For example, the screen display layout and operation menu may be automatically switched according to the value of the trust index and the evaluation stage, and the functions permitted to the target entity may be increased or decreased in stages. In addition, network bandwidth and usage time may be controlled, and the number of connections and transfer volume may be expanded for target entities that have obtained a high trust index, while communication speed and access timing may be restricted in the case of suspicious access. Furthermore, in cooperation with other linked services, extended operations may be implemented, such as automatically approving applications for new services when the trust index rises, or expanding the scope of use of payment functions and external API integration. In this way, by introducing a design that flexibly changes the range of functions and usage conditions according to the trust index, rather than simply granting or denying access, the balance between ensuring security and convenience in interactions with the target entity can be controlled more precisely.

[0089] (Variation 1) Furthermore, looking at the information processing system described above from a different perspective, the problem (objective) that the information processing system according to this embodiment aims to solve can also be understood as "to perform access control more strictly and flexibly using information obtained from the behavioral analysis of target entities."

[0090] Given the problem as described above, an invention as a means to solve the problem might look like this: "An information processing method executed by an information processing device, An information retrieval step to obtain information about the target entity, An analysis step in which the behavioral analysis of the target entity is performed based on the aforementioned information, An estimation step of estimating the confidence index of the target entity from the results information of the behavioral analysis, Information processing methods including [this].

[0091] With the above configuration, as shown in Figure 6, since trust indicators can be estimated based on the behavioral analysis of the target entity, it is possible to achieve more dynamic and practical security enhancements and highly accurate evaluations compared to conventional technologies that use only static account information and attribute information. Specifically, the information acquisition unit acquires information including behavioral data from the target entity, the analysis unit grasps what kind of operations and access patterns the target entity is taking in an analysis step, and the trust unit estimates trust indicators using the results, enabling flexible judgments according to different situations and time periods.

[0092] Furthermore, by extending the evaluation items in the above analysis step to multiple dimensions, it becomes easier to automatically assess the risk of access or operations that deviate from the target entity's normal behavioral patterns as higher, and to impose additional checks or restrictions as needed. For example, if access occurs from a remote location late at night, even though access is normally made from the same location during the day, the anomaly in the behavior can be captured, enabling highly accurate detection of fraud or impersonation.

[0093] Furthermore, the estimation step, which estimates trust metrics based on the behavioral analysis results of the target entity, offers significant advantages in enhancing adaptability to environmental changes. For example, instead of fixedly determining the permission level and access status of the target entity, dynamically updating trust metrics in accordance with the accumulation of behavioral data (access frequency, operation logs, usage time, etc.) makes it easier to maintain and improve operational efficiency and security levels in the long term while minimizing policy settings at the time of implementation.

[0094] Furthermore, when the data obtained in the information acquisition step of the present invention is combined with other authentication methods (such as biometric authentication or tokenization) or natural language-based inquiry processes, the results of trust index estimations based on behavioral analysis can be used as additional supporting information. By configuring such a multi-layered security approach, it is expected that a stronger defense can be built against threats and fraudulent activities that are difficult to prevent with conventional single authentication methods.

[0095] Furthermore, an approach that combines behavioral analysis and confidence metric estimation makes it possible to finely control access as needed. Specifically, it can automatically differentiate access by allowing only limited functions to target entities with low estimated confidence metrics, and gradually opening up access to highly sensitive information for those with high metrics. This leads to reduced operational costs and improved user convenience.

[0096] As described above, the information processing method related to Modification 1 is characterized by its ability to estimate dynamic and realistic reliability indicators based on behavioral analysis, and can meet security requirements and flexible access control needs that cannot be adequately addressed by conventional technologies. As it is a highly versatile technological foundation applicable to any business field, it can be used across a variety of system areas that seek a balance between security and convenience.

[0097] (Modification 2) Furthermore, if we consider the information processing system described above from a different perspective, the problem (objective) that the information processing system according to this embodiment aims to solve can also be understood as "to accurately estimate the reliability index of a target entity by utilizing the biometric authentication information of that target entity."

[0098] Given the problem as described above, an invention as a means to solve the problem might look like this: "An information processing method executed by an information processing device, An information retrieval step to obtain information about the target entity, An extraction step of extracting biometric authentication information related to the target entity from the aforementioned information, An estimation step for estimating the confidence index of the target entity from the extraction result information of the extraction step, Information processing methods including [this].

[0099] With the above configuration, as shown in Figure 7, the information acquisition unit acquires biometric authentication information from the target entity, extracts the identity and biometric features of the target entity from this information (extraction step in the extraction unit), and estimates a trust index based on the extracted information (estimation step in the estimation unit). Specifically, by combining technologies that support multiple biometric information, such as voiceprints and facial features, in addition to high-precision biometric authentication elements such as fingerprints and irises, it becomes easier to prevent unauthorized access and impersonation that are difficult to detect with static IDs and passwords. Furthermore, by using a model learned from a vast amount of historical data, even if the acquired biometric authentication information has changed slightly (due to aging, changes in physical condition, etc.), the system can determine identity while absorbing a certain degree of variability, thus offering high operational flexibility.

[0100] Furthermore, this technology is particularly effective when the target entity is a human being (including through the user terminal 14). Human biometric information is highly unique, and hardware infrastructure such as fingerprint sensors and cameras are readily available for everyday use, so incorporating the acquisition and authentication process enables highly accurate identity verification. On the other hand, as mentioned above, biometric information can fluctuate due to health conditions, etc., so it is advisable to set a certain tolerance range on the model side or to use additional auxiliary authentication (such as entering a passcode) in combination to balance usability and security.

[0101] Furthermore, even when the target entity is an agent, there are cases where a concept equivalent to biometric authentication information can be introduced. For example, if a group of sensors built into a physical robot or smart device can acquire unique "vital data" or "movement patterns," then this characteristic information can be used as an "analogy for biometric authentication" to manage agent-specific movement signatures and authentication data linked to the owner. This makes it easier to estimate trust indicators by leveraging the unique characteristics of each device and agent, even in environments where agents collaborate to perform tasks, and to achieve secure collaboration.

[0102] As described above, the integration with biometric authentication using Modification 2 is beneficial in complementing or enhancing normal password authentication and attribute-based access control. In particular, it is easy to implement and provides a high level of security when the target entity is a human, and even with mechanical entities such as agents, it can utilize "device fingerprints" or unique sensor information, thus having broad applicability. This is expected to continuously improve the accuracy of trust metric estimation, enable flexible access control according to the user's authentication level, and achieve both security assurance and convenience.

[0103] The "biometric authentication analogy" of the agent mentioned above refers to the concept of setting unique authentication data and characteristic information equivalent to "biometric authentication" for humans for agents that do not have a physical body, and using this to determine authenticity and grant authority to other agents. Specifically, it has the following components and operational forms.

[0104] (1) Acquisition of proprietary sensor information By utilizing various sensors (motion sensors, environmental sensors, location sensors, and even mechanical feature recognition sensors) on robots, smart devices, and AR / VR terminals equipped with AI agents, the unique motion patterns and response timings of the agents are acquired. While this may not have the absolute uniqueness of human fingerprints or iris data, combining multiple types of sensor information allows for a highly accurate estimation of the "agent's uniqueness."

[0105] (2) Learning the behavior signature The system observes and learns the different processing procedures, task execution speed, and power consumption behavior of each agent over a certain period, and then records and manages them as if they were "biological characteristics." For example, by comprehensively treating the time variation patterns when processing a specific computation task, the changes in CPU load, and the procedure for signing security certificates as "operational signatures," a mechanism can be built to eliminate impersonation and duplicate agents.

[0106] (3) Integration with verifiable hardware identifiers Recent devices include mechanisms that provide device-specific identifiers at the hardware level, such as TPM (Trusted Platform Module) and secure enclaves. By working in conjunction with these mechanisms, agents can encrypt and securely manage data corresponding to "vital information" within the agent. This provides attack resistance and authenticity equivalent to human biometric information, and can be considered "agent-specific electronic 'biometric information'."

[0107] (4) Inter-agent authentication and calculation of trust metrics In an environment where agents collaborate, the "vital data" and "operational signatures" held by each agent can be exchanged and verified to assess the risk of tampering and imitation. The results of this assessment are reflected as "trustworthiness" and "authority levels" through user approval or an orchestration system, and specific operations or API calls are permitted or restricted accordingly. By applying this analogy of biometric authentication to the analysis of relationships between agents, a mechanism can be established in which agents cooperate while mutually verifying each other's trustworthiness.

[0108] (5) Operational and implementation considerations The data equivalent to "biometric information" in an agent can take on a wider variety of forms than human biometric authentication, but ensuring its legitimacy and uniqueness is a crucial challenge. It may be advisable to incorporate a mechanism that can prove that it is the "same agent" within a certain margin, while allowing for environmental changes such as sensor failures, firmware updates, and changes in the cloud environment. Furthermore, since attackers may be able to replay operation logs and create virtual agents, it may be advisable to use encryption technology to prove authenticity and to complement the verification process using both online and offline methods.

[0109] (Variation 3) Furthermore, looking at the information processing system described above from a different perspective, the problem (objective) that the information processing system according to this embodiment aims to solve can also be understood as "achieving both security and convenience by adjusting the authentication strength and access scope as needed."

[0110] Given the problem as described above, an invention as a means to solve the problem might look like this: "An information processing method executed by an information processing device, An information acquisition step that obtains at least one of the information about the target entity and the request information from the target entity, A permission setting step to set access permissions for the target entity based on the aforementioned information, Information processing methods including [this].

[0111] According to Modification 3, as shown in Figure 8, the information acquisition unit acquires information about the target entity and request information from that target entity. The risk assessment unit then evaluates the risks of this information, and the authorization setting unit sets access permissions for the target entity according to the evaluation results. This configuration offers greater flexibility than conventional uniform control mechanisms, as it allows for dynamic adjustment of authentication strength and access scope according to the risk level of each request and the confidentiality of the requested information. In other words, for normal operations and low-risk requests, smooth access is permitted while minimizing the effort required for authentication. On the other hand, when highly confidential information or requests deemed high-risk are made, control can be strengthened by automatically applying additional authentication processes or gradual restrictions. This variability of control policies allows for operation that satisfies strict security requirements without unnecessarily complicating the processing flow. As a result, not only is the overall system security enhanced, but improvements in convenience and operational efficiency can also be expected. Furthermore, since the access permission setting step is based on the target entity's own information and request content, it becomes easier to build control logic that appropriately meets requirements even in environments where AI agents collaborate with each other and where human users coexist. As described above, by flexibly switching authentication strength and access scope as needed, the present invention offers a level of usefulness not found in conventional access control methods, as it achieves a high degree of balance between security and convenience.

[0112] In this specification, the step of describing a program to be recorded on a recording medium includes not only processes that are performed chronologically in that order, but also processes that are not necessarily performed chronologically, but are executed in parallel or individually. Furthermore, in this specification, the term "system" refers to an overall system composed of multiple devices, means, etc.

[0113] The above describes implementations where the target entity is an AI agent or a user, but the target entity is not limited to these. For example, it can be applied to authentication between delivery boxes and automated delivery vehicles, authentication for sending and receiving childcare notebooks when a child leaves a nursery school, authentication requiring special handling based on on-site judgment (so-called VIP pass type), and authentication during inheritance of estates and personal belongings. Examples of each function are described below.

[0114] (1) Authentication between delivery boxes and automated delivery vehicles In a configuration where both the delivery box and the automated delivery vehicle are treated as agents and bidirectional authentication is performed, not only can a "delivery status acquisition module" and a "location-based inquiry step" be introduced, in addition to the simple exchange of PINs. Specifically, the delivery box dynamically grasps the automated delivery vehicle's route and estimated arrival time, and the inquiry step verifies whether it matches the actual arrival time and location information. By reflecting these verification results (consistency, delay status, etc.) in a reliability index and automatically determining whether or not to control key issuance, a safer and more flexible package delivery becomes possible.

[0115] Furthermore, other concrete examples include not only authentication between "delivery boxes and automated delivery vehicles," but also the construction of authentication and information verification processes between human users and automated delivery vehicles (agents). For example, imagine a scenario where an automated delivery vehicle is transporting food and drinks such as beef bowls to a cherry blossom viewing venue, and a user requests to pick up the package midway along the route. In this case, if the user inquires something along the lines of, "I'm [username], and I told them I would pick up the package midway. Is it possible to receive it?", the agent on the automated delivery vehicle side could dynamically execute an authentication or authorization verification process based on the inquiry information to determine whether the user is a legitimate user. Specifically, the automated delivery vehicle could obtain "whether or not it has been contacted by the person in charge in advance" as inquiry information, and through a trust index estimation step that comprehensively evaluates the content of the user's response (orderer name, order details, etc.), it could perform the handover if it is confirmed to be appropriate, or temporarily refuse the handover if the information is unclear or confusing. For example, when a user asks, "Can I pick up one item along the way?", the automated delivery vehicle's agent checks the consistency of the person in charge's information and the order details, and based on the results, responds with something like, "We have confirmed it, so we will hand it over," or "We have not heard about this and it is unclear, so we cannot hand it over."

[0116] By incorporating this inquiry and response process between "humans and automated delivery vehicles (agents)," appropriate party authentication can be implemented even when dynamically changing delivery destinations or handover locations, which were previously set uniformly. As a result, it becomes possible to maintain user convenience while preventing fraudulent receipt and information confusion. Similar to the aforementioned example of delivery boxes and automated delivery vehicles, a key feature of this embodiment is that the flow for estimating reliability indicators based on the generation of inquiry information and analysis of response information can function universally.

[0117] (2) Authentication of sending and receiving childcare notebooks (management of transfers and withdrawals at nurseries, kindergartens, etc.) In the context of exchanging childcare notebooks, assuming a scenario where parents, childcare workers, and the nursery's system communicate on the same network, an "inquiry step" can be incorporated into the authentication modules of each parent and nursery staff member. This allows for the analysis of the consistency of responses and the status of required information input, enabling the updating of reliability indicators. For example, by repeatedly inquiring about a child's detailed medical history and length of enrollment, and comprehensively evaluating the responses, it is possible to implement a system that ensures access to childcare notebooks and the procedures for withdrawal are handled reliably.

[0118] (3) Authentication requiring special handling (so-called VIP pass) In situations requiring rapid and rigorous authentication that differs from the normal flow, such as emergency situations or VIP handling, the system can be configured with an "emergency / exception mode control unit" that switches the content of inquiry steps and evaluation criteria according to the mode change. For example, if the target entity requesting a VIP pass can provide a response demonstrating high reliability indicators in a short time, an exception can be granted, while if the response is significantly inconsistent, rigorous additional verification can be performed. This creates a system that automatically switches between simple and strict authentication according to the requirements.

[0119] (4) Authentication when inheriting estates and personal belongings Inheritance procedures typically involve multiple heirs and legal professionals. By applying this embodiment to incorporate a "multi-stakeholder inquiry subsystem," inquiry information can be presented to each heir in stages, making it easier to automatically detect inconsistencies and contradictions in their responses. Furthermore, by incorporating a multi-stage flow that only permits electronic signatures and meeting minute management when a sufficiently high confidence index is obtained, fraudulent declarations and signature problems can be minimized.

[0120] (5) Versatility of this embodiment As described above, the inquiry steps and evaluation modules can be optimized according to the specific challenges of each application field (such as key handover, secure management of childcare information, policy switching in emergencies, and consistency of agreements among multiple parties), including delivery services, childcare operations, special cases, and inheritance-related matters. Based on the core processes of this embodiment, namely "generating inquiries to target entities," "acquiring and analyzing response information," and "estimating trust indicators," the system can be easily implemented in different scenarios simply by adding domain-specific functional modules and operating interfaces. As a result, the information processing system disclosed herein can be deployed as a technological foundation that enhances security, reliability, and flexibility in a wide variety of situations, going beyond the framework of simple server-agent authentication.

[0121] <Note> This embodiment includes the following disclosures.

[0122] (Note 1) An information processing method performed by an information processing device, An information retrieval step to obtain information about the target entity, A query step that generates query information for estimating the reliability index of the target entity from the aforementioned information and presents the said query information to the target entity, An estimation step of obtaining response information from the target entity to the aforementioned inquiry information and estimating the confidence index of the target entity from said response information, Information processing methods including

[0123] (Note 2) The aforementioned inquiry information includes information expressed in natural language. The information processing method described in Appendix 1.

[0124] (Note 3) The information relating to the target entity includes the output information history of the target entity. The information processing method described in Appendix 1 or Appendix 2.

[0125] (Note 4) The estimation step estimates the confidence index based on the relationship between the inquiry information and the response information. The information processing method described in any one of the items in Appendix 1 to Appendix 3.

[0126] (Note 5) The aforementioned query information is information that prompts the target entity to perform a process, The aforementioned response information is information relating to the processing result of the target entity. The information processing method described in any one of the items in Appendix 1 to Appendix 4.

[0127] (Note 6) The process includes a permission setting step of setting access permissions to the target entity in accordance with the aforementioned trust indicators, The information processing method described in any one of the items in Appendix 1 to Appendix 5.

[0128] (Note 7) In an information processing device, An information retrieval step to obtain information about the target entity, A query step that generates query information for estimating the reliability index of the target entity from the aforementioned information and presents the said query information to the target entity, An estimation step of obtaining response information from the target entity to the aforementioned inquiry information and estimating the confidence index of the target entity from said response information, A program for executing an information processing method that includes this method.

[0129] (Note 8) An information processing system executed by an information processing device, An information acquisition unit that acquires information about the target entity, A query unit generates query information for estimating the reliability index of the target entity from the aforementioned information and presents the query information to the target entity. An estimation unit that obtains response information from the target entity to the inquiry information and estimates the confidence index of the target entity from the response information, An information processing system having [Explanation of Symbols]

[0130] 10. Information Processing Systems 12 servers (information processing devices) 14. User terminals (information processing devices) 16 Target Entity (Target Entity)

Claims

1. An information processing method performed by an information processing device, An information retrieval step to obtain information about the target entity, A query step that generates query information for estimating the reliability index of the target entity from the aforementioned information and presents the said query information to the target entity, An estimation step of obtaining response information from the target entity to the inquiry information and estimating the confidence index of the target entity from said response information, Information processing methods including

2. The aforementioned inquiry information includes information expressed in natural language. The information processing method according to claim 1.

3. The information relating to the target entity includes the output information history of the target entity. The information processing method according to claim 1.

4. The estimation step estimates the confidence index based on the relationship between the inquiry information and the response information. The information processing method according to claim 1.

5. The aforementioned query information is information that prompts the target entity to perform a process, The aforementioned response information is information relating to the processing result of the target entity. The information processing method according to claim 1.

6. The process includes a permission setting step of setting access permissions to the target entity in accordance with the aforementioned trust indicators, The information processing method according to claim 1.

7. In an information processing device, An information retrieval step to obtain information about the target entity, A query step that generates query information for estimating the reliability index of the target entity from the aforementioned information and presents the said query information to the target entity, An estimation step of obtaining response information from the target entity to the inquiry information and estimating the confidence index of the target entity from said response information, A program for executing an information processing method that includes this method.

8. An information processing system executed by an information processing device, An information acquisition unit that acquires information about the target entity, A query unit generates query information for estimating the reliability index of the target entity from the aforementioned information and presents the query information to the target entity. An estimation unit that obtains response information from the target entity to the inquiry information and estimates the confidence index of the target entity from the response information, An information processing system having [a certain feature].

Images