Artificial Intelligence Model-Interfaced Data Governance and Audit Log Management System and Method

Abstract

The objective of the present invention is to prevent indiscriminate data access and bypassing of permissions in advance by controlling data requests generated by an artificial intelligence model or agent to be allowed, restricted, or suspended based on policies through a standardized integration interface before they are executed on an external data server. A data governance and audit log management system based on an artificial intelligence model linkage interface according to the present invention for solving the aforementioned problem comprises: a standardized linkage interface unit for receiving a natural language-based query from an artificial intelligence model or agent and / or a data request generated based on said natural language-based query; a governance control unit for controlling the execution of said data request to allow, restrict, or suspend according to a pre-set data access policy before said data request is executed on an external data server; and an audit log management unit for generating and storing an audit log for said data request, which includes at least one of request subject information, request content, and a judgment result of said governance control unit.

Description

Technology Field

[0001] The present invention relates to a data governance and audit log management system and method, and more specifically, to the field of AI-linked data access control and data governance technology that controls data requests generated by an AI model or agent based on policies before they are executed on an external data server and manages the processing process as an audit log. Background Technology

[0002] Systems in which artificial intelligence models, such as large-scale language models (LLM), or agents driven by them interpret natural language-based queries to automatically generate and execute data requests to external data servers, such as databases (SQL), data warehouses, and internal APIs, are widely used.

[0003] In particular, NL2SQL agents, tool-using agents, and RAG-based query systems can significantly increase business productivity, but since their operation is based on an automation pipeline leading from “natural language → executable data request,” the creation, delivery, and execution of data requests are performed through various components (models, agent runtimes, plugins / connectors, gateways, DB engines, etc.) and may be connected to external data servers in different ways depending on the service environment.

[0004] As a result, a structure is likely to form where multiple AI models / agents within the same organization access external data servers through separate connectors or integration specifications, and unless a single path (standardized integration interface) that can be centrally controlled and audited is established, data access may become distributed and inconsistent.

[0005] However, conventional technology had limitations in consistently determining whether to allow, restrict, or hold data requests generated by artificial intelligence models / agents based on policies at the pre-execution stage, and in integrating and recording the grounds for such judgments and pre- and post-execution information in an auditable format.

[0006] For example, existing DB permission management, API gateways, network firewalls, or application-specific logs each provide only partial control or recording, resulting in problems such as (i) difficulty in proactively blocking agents from bypassing permission scopes or generating queries with excessive scope, (ii) issues where execution results are only recognized retrospectively even if they cause excessive resource consumption or skyrocketing costs, (iii) difficulty in reproducible auditing because the basis for policy judgments regarding “who / what / why allowed or denied” (reason codes, policy versions / snapshots, etc.) is not retained, and (iv) difficulty in systematically tracking and controlling repeated access to similar or repetitive queries (including retries with only changed wording). Prior art literature

[0007] Republic of Korea Published Patent Application No. 10-2025-0072511 (Publication Date: May 23, 2025) Republic of Korea Published Patent Application No. 10-2025-0154310 (Publication Date: October 28, 2025) The problem to be solved

[0008] The objective of the present invention for solving the aforementioned problem is to prevent indiscriminate data access and authorization bypassing in advance by controlling data requests generated by an artificial intelligence model or agent to be allowed, restricted, or suspended based on policies through a standardized integration interface before they are executed on an external data server.

[0009] Furthermore, the objective of the present invention is to provide transparency and a reproducible audit system regarding data access by predicting the risk and resource consumption costs based on the structural characteristics of data requests prior to execution, and by integrating the basis for such judgment and pre- and post-execution information into an audit log.

[0010] Furthermore, the objective of the present invention is to improve data governance and compliance response capabilities by identifying and controlling repeated access to the same or substantially the same data requests, and by systematically analyzing and managing data access behaviors based on artificial intelligence models. means of solving the problem

[0011] A data governance and audit log management system based on an artificial intelligence model linkage interface according to the present invention for solving the aforementioned problem comprises: a standardized linkage interface unit for receiving a natural language-based query or a data request generated based on said natural language-based query from an artificial intelligence model or agent; a governance control unit for controlling the execution of said data request to allow, restrict, or suspend according to a pre-set data access policy before said data request is executed on an external data server; and an audit log management unit for generating and storing an audit log including at least one of request subject information, request content, and a judgment result of said governance control unit for said data request.

[0012] In addition, in a data governance and audit log management system based on an artificial intelligence model linkage interface according to the present invention, the governance control unit predicts the risk of data exposure or resource consumption costs resulting from the execution of the data request based on the structural characteristics of the data request, and determines whether to execute the data request by considering the prediction result together with the data access policy.

[0013] In addition, in the data governance and audit log management system based on an artificial intelligence model linkage interface according to the present invention, the audit log management unit is characterized by including and storing policy information or policy identification information regarding a data access policy applied to the data request in the audit log, thereby enabling subsequent reproduction of the judgment result regarding the data request.

[0014] In addition, in a data governance and audit log management system based on an artificial intelligence model linkage interface according to the present invention, the audit log management unit is characterized by storing the execution result of the data request and the time or cost incurred for execution in the audit log when the data request is executed.

[0016] In addition, the method for managing data governance and audit logs based on an artificial intelligence model linkage interface according to the present invention for solving the aforementioned problem comprises: (a) receiving a natural language-based query from an artificial intelligence model or agent or a data request generated based on said natural language-based query; (b) determining whether to allow, restrict, or suspend the execution of said data request according to a pre-set data access policy before said data request is executed on an external data server; (c) transmitting said data request to said external data server for execution, blocking transmission, or suspending processing according to said determination result; and (d) generating and storing an audit log for said data request that includes at least one of request subject information, request content, determination result, and execution result.

[0017] In addition, in the method for data governance and audit log management based on an artificial intelligence model linkage interface according to the present invention, the step (b) is characterized by including a step of predicting the risk or resource consumption cost associated with the execution of the data request based on the structural characteristics of the data request.

[0018] In addition, in the method for data governance and audit log management based on an artificial intelligence model linkage interface according to the present invention, the step (d) is characterized by including and storing information indicating the reason for determining whether to execute the data request in the audit log.

[0019] In addition, in the method for data governance and audit log management based on an artificial intelligence model linkage interface according to the present invention, if the processing of the data request is suspended in step (c), the data request is re-evaluated according to additional approval or policy change.

[0020] In addition, the method for data governance and audit log management based on an artificial intelligence model linkage interface according to the present invention is characterized by comprising, in step (a), a step of preprocessing a database query generated based on the data request to include a normalized representation or identification information for identifying its structural characteristics, and in step (d), a step of storing the normalized representation or identification information together in the audit log so as to enable repetitive access control and post-tracing for the same or substantially the same data request.

[0022] In addition, the computer program according to the present invention for solving the above-described problem is characterized by being stored in a storage medium to execute a data governance and audit log management method based on the above-described artificial intelligence model linkage interface on a computer. Effects of the invention

[0023] According to the present invention, unauthorized access and data leakage can be effectively prevented by controlling data requests generated by an artificial intelligence model or agent on a policy basis before they are executed on an external data server.

[0024] In addition, according to the present invention, data access paths of multiple artificial intelligence models and agents can be unified through a standardized interoperability interface, thereby enabling consistent access control without governance gaps.

[0025] In addition, according to the present invention, by predicting the risk and resource consumption costs based on the structural characteristics of data requests before execution, excessive system load and skyrocketing costs can be prevented in advance.

[0026] In addition, according to the present invention, the results of the judgment regarding the permission, restriction, or suspension of data requests and the reasons thereof are recorded in an audit log, thereby improving the transparency and explainability of the data access decision process.

[0027] In addition, according to the present invention, by integrally recording pre-execution information, execution results, time required, and costs, it is possible to facilitate post-audit and regulatory compliance regarding data access.

[0028] In addition, according to the present invention, repeated access and attempts to bypass policies can be effectively blocked by identifying identical or substantially identical data requests using the normalized representation or identification information of a query.

[0029] In addition, according to the present invention, business continuity can be maintained while preventing immediate execution of high-risk requests through a hold and re-evaluation mechanism.

[0030] In addition, according to the present invention, past judgments can be reproduced even after a policy change through a policy snapshot-based judgment record, thereby strengthening dispute response and internal control.

[0031] In addition, according to the present invention, data access behavior based on artificial intelligence models can be systematically analyzed to continuously improve governance policies and access rules.

[0032] In addition, according to the present invention, data security, system stability, cost management, and compliance response can be achieved simultaneously within a single framework, thereby improving the overall reliability of the artificial intelligence utilization environment. Brief explanation of the drawing

[0033] FIG. 1 is a diagram illustrating the block configuration of a data governance and audit log management system based on an artificial intelligence model linkage interface according to an embodiment of the present invention. FIG. 2 is a schematic diagram of a connection illustrating a structure in which a data governance and audit log management system based on an artificial intelligence model linkage interface according to an embodiment of the present invention mediates, controls, and records data requests between an artificial intelligence agent and a database (DB). FIG. 3 is a flowchart showing the detailed flow of a data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating the specific step-by-step flow of a data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention. Specific details for implementing the invention

[0034] Before describing the present invention in detail, it should be understood that the terms and words used in this specification should not be interpreted as being limited to their ordinary or dictionary meanings, and that the inventor of the present invention may appropriately define and use the concepts of various terms to best describe their invention, and furthermore, that these terms and words should be interpreted in a meaning and concept consistent with the technical spirit of the present invention.

[0035] In other words, it should be understood that the terms used in this specification are used merely to describe preferred embodiments of the present invention and are not intended to specifically limit the content of the present invention, and that these terms are defined in consideration of various possibilities of the present invention.

[0036] In addition, it should be noted that in this specification, singular expressions may include plural expressions unless the context clearly indicates a different meaning, and that even if they are expressed in a similarly plural form, they may include the meaning of the singular.

[0037] Throughout this specification, where it is stated that a component "includes" another component, unless specifically stated otherwise, this may mean that it does not exclude any other component but may include any other component.

[0038] Furthermore, it should be noted that in cases where it is stated that a component "exists inside or is installed in connection with" another component, this component may be installed in direct connection or contact with the other component, or it may be installed at a certain distance apart, and in the case where it is installed at a certain distance apart, there may be a third component or means for fixing or connecting the component to the other component, and a description of this third component or means may be omitted.

[0039] On the other hand, if it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there is no third component or means.

[0040] Likewise, other expressions describing the relationship between each component, such as “between” and “right between”, or “adjacent to” and “directly adjacent to”, should be interpreted as having the same intent.

[0041] In addition, it should be understood that in this specification, terms such as “one side,” “other side,” “one side,” “other side,” “first,” “second,” etc., are used to clearly distinguish one component from another component, and that the meaning of the component is not restricted by such terms.

[0042] In addition, position-related terms such as "up," "down," "left," and "right" used in this specification should be understood as indicating the relative position of the corresponding component in the drawing, and unless an absolute position is specified, these position-related terms should not be understood as referring to an absolute position.

[0043] Furthermore, in specifying the reference numerals for each component of each drawing in this specification, the same component has the same reference numeral even if it is shown in different drawings; that is, the same reference numeral throughout the specification indicates the same component.

[0044] In the drawings attached to this specification, the size, position, connection relationships, etc., of each component constituting the present invention may be described in a partially exaggerated, reduced, or omitted manner for the convenience of explanation or to sufficiently clearly convey the concept of the present invention, and therefore, the proportions or scale may not be strictly accurate.

[0045] In addition, in the following description of the present invention, detailed descriptions of components, such as prior art and known technology, that are deemed to unnecessarily obscure the essence of the present invention may be omitted.

[0047] Preferred embodiments of the present invention will be described in detail below with reference to the drawings.

[0048] FIG. 1 is a diagram illustrating the block configuration of a data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to an embodiment of the present invention.

[0049] The data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to an embodiment of the present invention is intended to solve problems of data leakage, unauthorized access, excessive resource consumption, and lack of audit traceability that may occur when data requests (e.g., SQL queries, API calls, etc.) generated by an artificial intelligence model or agent from natural language-based queries are executed indiscriminately on an external data server.

[0050] That is, the governance and audit log management system according to an embodiment of the present invention secures data governance by controlling execution to allow, restrict, or suspend it according to a pre-set data access policy at a stage prior to the execution of a data request on an external data server, and simultaneously generates and stores audit logs regarding the requesting entity, the content of the request, and the policy judgment result, thereby providing a data governance and audit log management system (100) based on an artificial intelligence model linkage interface that enables post-audit and compliance response.

[0051] The key terms used in the present specification below may be defined as follows.

[0052] An “artificial intelligence model” refers to a computational entity that generates data requests or is involved in the generation of data requests in response to natural language-based queries or input data, and may include various machine learning-based models such as large-scale language models (LLM).

[0053] In addition, “agent” refers to a software component that configures and transmits data requests based on the output or inference results of an artificial intelligence model, and may include, for example, an NL2SQL agent, a tool-calling agent, etc.

[0054] "Natural language-based query" refers to input that expresses a user's question or instruction in natural language.

[0055] “Data request” refers to an executable request generated based on a natural language-based query, and may include, for example, SQL, API requests, search requests, data extraction / aggregation requests, etc. that can be processed by an external data server.

[0056] “External data server” is a system that is the target of execution for the above data request, and may include a database, data warehouse, data lake, analysis engine, API server, etc.

[0057] “Data access policy” refers to a set of rules for determining whether to execute a data request or the conditions for execution, and may include conditions such as the authority and role of the requesting entity (user / agent), the scope of data resources (tables / columns / resources), the request type (query / update), whether sensitive information is included, time zones, and usage limits.

[0058] “Allow, Restrict, or Hold” means approving the execution of a data request, blocking execution or restricting certain conditions, or suspending execution until further approval or review is completed, respectively.

[0059] “Audit log” refers to a record of the data request processing process and includes at least one of the requester information, the request content, and the policy judgment result.

[0061] More specifically, as shown in FIG. 1, the data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to an embodiment of the present invention may be configured to include a standardized linkage interface unit (110), a governance control unit (130), and an audit log management unit (150).

[0062] Here, the standardized interoperability interface unit (110) is configured to receive a natural language-based query and / or a data request generated based on the natural language-based query from an artificial intelligence model or agent, and may be configured to include an interface providing module (111), a request receiving and validation module (112), a normalization and structuring module (113), and a request context generating module (114).

[0063] The interface providing module (111) can provide a communication interface for an artificial intelligence model or agent to transmit a data request to the system.

[0064] An interface applied in one embodiment of the present invention may be implemented using HTTP-based APIs, RPCs, message queues, or socket communication, and standardizes the message format of requests to enable data requests to be transmitted in the same specifications even to different artificial intelligence models / agents. Accordingly, the system can perform consistent policy application and audit log recording for various agent implementations.

[0065] In addition, the interface providing module (111) described above is a functional configuration that provides a standardized communication means for an artificial intelligence model or agent to interact with an external data server, and according to one embodiment, may include a Model Context Protocol (MCP) based interaction module (hereinafter “MCP module”).

[0066] Here, the term MCP module refers to an integration module that enables an artificial intelligence model or agent to generate and transmit a data request by structuring the request format, context information, authentication information, and tool call rules according to a certain protocol specification.

[0067] This MCP module can provide a technical basis for enabling the governance control unit (130) and the audit log management unit (150) to perform policy judgment and log recording based on consistent standards by converting and relaying natural language-based queries generated by an artificial intelligence model or agent and data requests based thereon into a standard message form that can be processed by the data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to an embodiment of the present invention.

[0068] That is, the MCP module applicable as an embodiment of the present invention functions as a standardized interface layer between an artificial intelligence agent and the system of the present invention, and can perform the role of converging data requests into a controllable form regardless of the internal implementation or learning method of the artificial intelligence model.

[0069] Meanwhile, the interface providing module (111) in the present invention is not necessarily limited to an MCP module and may include various interlocking modules that can perform the same or substantially the same function as an MCP module.

[0070] For example, an API gateway module that standardizes and transmits tool call requests from an AI agent in the form of HTTP-based API calls, an RPC-based integration module that relays data requests according to remote procedure call specifications, a message broker-based integration module that asynchronously transmits data requests using message queues or event streams, or an agent runtime embedded integration module included within the AI ​​agent's execution environment that intercepts data requests and converts them into a standard format may be used.

[0071] Although these modules may differ in implementation methods or communication standards, they perform the same technical function as the MCP module in that they transmit data requests generated from artificial intelligence models or agents in a standardized form that the system of the present invention can interpret, control, and audit; therefore, the present invention should be understood as a concept that encompasses all of these various interconnected modules.

[0072] And, as shown in FIG. 1, the request reception and validation module can check whether the request received through the integration interface includes required fields, whether the format of the data request (e.g., SQL syntax, API endpoint specification, etc.) is valid, and whether it includes an abnormal payload (e.g., excessive length, prohibited command pattern, etc.).

[0073] This module provides technical effects that prevent external data server errors caused by formal errors in advance and reduce the inflow of malicious or abnormal requests into the policy judgment stage.

[0074] The request normalization and structuring module (113) normalizes natural language-based queries and / or data requests into a form suitable for policy judgment.

[0075] For example, in the case of SQL requests, superficial differences such as spaces, case sensitivity, comments, and aliases can be removed, and structural information such as the query type (SELECT / UPDATE / DELETE, etc.), target table / column, WHERE clause, and join structure can be extracted.

[0076] Such normalization and structuring enable consistent policy application for identical or substantially identical requests and can be used as input for request characteristic-based risk / resource consumption assessments.

[0077] The request context creation module constructs a “request context” by combining request subject information (e.g., user ID, agent ID, authentication token), request content (natural language query / data request), and metadata such as the request time.

[0078] The request context subsequently becomes the unit of judgment for which the governance control unit (130) applies data access policies, and can be used as a basic data structure referenced when the audit log management unit (150) generates audit logs.

[0079] And, as shown in FIG. 1, the governance control unit (130) is configured to control the execution of a data request according to a pre-set data access policy before the data request is executed on an external data server, and may be configured to include a policy storage and loading module (130), a request subject authentication module (132), a policy evaluation module (134), an execution decision orchestration module (135), and a hold processing linkage module (136).

[0080] The access control of such a governance control unit (130) can perform role-based access control (RBAC) functions.

[0081] In other words, through Role-Based Access Control (RBAC), access rights to specific tables, columns, or query types (such as SELECT / INSERT) can be finely controlled based on user roles, allowing only approved queries to pass through.

[0082] More specifically, the policy storage and loading module (130) may be configured to store a pre-set data access policy and load the policy at the time of judgment to provide it for policy evaluation.

[0083] In one embodiment, the policy may include role-based rules, table / column-based rules, request type-based rules, sensitive information rules, usage / cost limit rules, etc., and may be configured so that an administrator can update the policy.

[0084] The request subject authentication module (132) can identify and authenticate the request subject by verifying credentials (e.g., tokens) included in the request context. For example, it can verify the validity of the token by linking with an external authentication server (Auth Server) or an internal authentication module and obtain the request subject's user ID / agent ID / role information.

[0085] Such a request subject authentication module (132) can prevent governance control from being neutralized by “anonymous requests” and enable authority-based control.

[0086] The policy evaluation module (134) evaluates whether a data request is executable by comparing the request context and the data access policy.

[0087] For example, a specific role can be configured to allow only queries on specific tables and restrict update requests, and requests containing specific sensitive columns can be configured to be restricted or held in abeyance.

[0088] Policy evaluation can be performed not only by a single rule but also by a combination of multiple rules, and in the event of conflicting rules, judgment can be made based on priority.

[0089] And, the execution decision orchestration module (135) determines the data request as one of “allow, restrict, or hold” based on the policy evaluation result.

[0090] Allow means to forward the data request to an external data server to enable execution, restrict means to block the execution of the data request or to reduce the scope of processing the request as necessary, and hold means to suspend the processing of the request until further review or approval is completed.

[0091] This module can structurally block indiscriminate access from external data servers by controlling the flow of data requests in the pre-execution phase.

[0092] And, the hold processing linkage module (136) may be configured to store data requests determined to be held in a separate waiting queue, or to send a review request to an approval entity (e.g., security officer, data governance manager), and to perform re-evaluation or execution based on the approval result.

[0093] This allows business continuity to be maintained while preventing the immediate execution of high-risk requests or requests requiring policy judgment.

[0094] In addition, the governance control unit (130) can perform a direct DB access blocking function that fundamentally blocks a user or agent from directly accessing the database (DB) and mediates communication only through a standardized integration interface or MCP server.

[0095] In addition, in the data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to the present embodiment, the governance control unit (130) may further include a risk and cost prediction module (133) that predicts, before execution, the risk of data exposure or resource consumption costs that may occur when the data request is executed on an external data server, based on the structural characteristics of the data request received from the artificial intelligence model or agent.

[0096] Here, structural characteristics may include, for example, the type of query for the data request (retrieval, update, etc.), the range of the target table and column, the condition clause and join structure, whether aggregate operations are included, etc.

[0097] In addition, these structural characteristics can be used as a basis for pre-determining the impact of data request execution results on system resources or the possibility of sensitive data exposure.

[0098] That is, by considering these prediction results of the risk and cost prediction module (133) of the governance control unit (130) together with the preset data access policy, it is possible to comprehensively determine whether to allow, restrict, or withhold the execution of a data request.

[0099] Accordingly, according to an embodiment of the present invention, rather than a reactive approach where a problem occurs after a data request is actually executed on an external data server, preemptive control considering risk and cost is possible at a pre-execution stage.

[0100] In particular, data requests generated by artificial intelligence models or agents are often difficult to intuitively predict in advance; however, in the embodiments of the present invention, by performing a predictive judgment based on the structural characteristics of the data request, problems such as excessive resource consumption or the massive exposure of sensitive information can be effectively prevented.

[0101] As a result, it provides a technical effect that can simultaneously ensure the safety of data access and the stable operation of system resources.

[0102] Meanwhile, as shown in FIG. 1, the audit log management unit (150) may be configured to generate and store an audit log including at least one of the requesting entity information, the request content, and the judgment result of the governance control unit (130) for a data request.

[0103] More specifically, the audit log management unit (150) may be configured to include a log schema definition and creation module (151), a log storage module (152), a log integrity protection module (153), and a log inquiry and reporting module (154).

[0104] Here, the log schema definition and generation module (151) can define the field configuration of the audit log and generate audit log records from the request context and policy judgment result.

[0105] The audit log includes at least one of the requesting entity information, the request content, and the judgment result of the governance control unit (130), and depending on the embodiment, may include additional information such as the request time, policy identification information, reason for hold, reason for restriction, etc.

[0106] The log storage module (152) may be configured to store the generated audit logs in an audit log storage. The storage may be implemented as a database, a distributed log storage system, a file-based storage, etc., and the stored logs may be used as grounds for post-audit, compliance checks, and abnormal access detection.

[0107] The log integrity protection module (153) can perform a function to prevent the audit log from being tampered with after the fact. In one embodiment, the integrity of the log can be verified by generating a hash for the audit log records and connecting them sequentially (hash chain), or by applying signature-based verification. This integrity protection provides a technical effect that strengthens the evidentiary value of the audit log.

[0108] The log inquiry and reporting module (154) can query logs stored in the audit log repository to report access history, policy violation history, status of pending / restriction occurrences, etc., by user / agent. Through this, the operator can monitor the data access behavior of the artificial intelligence model / agent and check the appropriateness of the governance policy.

[0109] In addition, in the data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to an embodiment of the present invention, the audit log management unit (150) may be configured to store policy information or policy identification information regarding the applied data access policy together in the audit log when determining whether to execute a data request.

[0110] In other words, rather than simply recording the processing results of data requests, information that identifies which policy criteria were used to make such a judgment can be logged.

[0111] Here, policy identification information can be implemented as the policy version, policy ID, reference information for the policy rule set, etc., which allows the policy status at the time the data request is processed to be clearly identified.

[0112] With such a configuration, the embodiment of the present invention provides a technical effect that enables the judgment result of a data request to be reproduced retrospectively.

[0113] In other words, since it is possible to trace the reason why a specific data request was allowed, restricted, or withheld—not merely as a result, but specifically “because a certain policy was applied”—it is possible to provide objective grounds for judgment in post-audits, internal control checks, regulatory responses, or the occurrence of disputes.

[0114] In particular, even in environments where data access policies change over time, the results of policy application at past points in time can be clearly verified, which can significantly enhance the reliability and transparency of data governance.

[0115] And, as shown in FIG. 1, the audit log management unit (150) applied to the data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to an embodiment of the present invention may be configured to store the time or cost of execution taken for the execution along with the execution result of the data request in the audit log when the data request is actually executed on an external data server.

[0116] The execution results may include whether the data request was successful, whether an error occurred, etc., and the time taken for execution may include query execution time, response delay time, etc.

[0117] In addition, execution costs can be calculated based on data server resource usage, throughput-based costs, and billing information in a cloud environment.

[0118] Accordingly, embodiments of the present invention provide the effect of being able to integrally record not only prior judgment information regarding data requests but also subsequent execution information.

[0119] This allows for the quantitative identification of the actual load and costs incurred by data requests on the system, and can be utilized for improving future data access policies, establishing cost control standards, or analyzing the request behavior of AI models / agents.

[0120] In particular, by systematically managing execution results and costs through audit logs, accountability and transparency regarding data access can be enhanced, and management efficiency in terms of system operation and compliance can be significantly improved.

[0121] Thus, the data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to an embodiment of the present invention receives natural language-based queries or data requests flowing in from an artificial intelligence model / agent in a unified manner through a standardized interface, and before the data request is executed on an external data server, the governance control unit (130) can apply a data access policy to allow, restrict, or postpone execution.

[0122] In addition, in the process, the audit log management unit (150) generates and stores an audit log that includes the requesting entity, the content of the request, and the policy judgment result, thereby enabling the technical effect of simultaneously securing pre-control and post-audit capabilities in an AI-based data access environment.

[0123] In this way, the audit log management unit (150), which is a component of the model linkage interface-based data governance and audit log management system (100) according to an embodiment of the present invention, can perform the following functions.

[0124] 1) Generate logs before and after query execution: Natural language queries received from the NL2SQL Agent and the AI ​​Agent, converted SQL, execution agent (User ID), access time, etc. can be recorded.

[0125] 2) Integrated record of execution results and meta-information: Query execution results (whether successful or unsuccessful), time taken for execution, and information on predicted / actual query execution costs can be integrated and stored in the audit log.

[0126] 3) Storage in a log repository: Accumulated audit logs can be stored in a secure internal data repository to prevent tampering and to manage them for use in compliance reviews.

[0128] FIG. 2 is a schematic diagram of a connection illustrating a structure in which a data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to one embodiment of the present invention mediates, controls, and records data requests between an artificial intelligence agent and a database (DB).

[0129] As shown in FIG. 2, an NL2SQL agent (10) driven by an artificial intelligence model transmits an SQL query or a corresponding data request generated based on a natural language-based query to a system (100) according to an embodiment of the present invention.

[0130] A system (100) according to an embodiment of the present invention receives the data request through a standardized interlocking interface unit (110) and transmits the received request to a governance control unit (130).

[0131] The governance control unit (130) can be linked with an external authentication server (Auth Server) to perform token-based authentication and role-based access control, thereby verifying the identity and authority of the requesting subject.

[0132] Additionally, the governance control unit (130) determines whether to allow, restrict, or suspend the execution of data requests according to the data access policy, and controls the transmission of only approved SQL queries to the downstream while blocking direct access to the database (20).

[0133] Meanwhile, data requests allowed to be executed by the governance control unit (130) are transmitted to the database (20) through the audit log management unit (150) and SQL queries are executed.

[0134] The audit log management unit (150) logs query content, request subject information, etc., in the pre-execution stage of a data request, and integrates metadata such as execution success, time taken, resource usage, or cost for execution result data returned from the database (20) and stores it as an audit log.

[0135] The audit log generated in this way is configured to enable post-auditing and tracing by including information both before and after the execution of the data request.

[0136] In addition, the execution result returned from the database (20) is transmitted to the NL2SQL agent (10) via the audit log management unit (150) and returned as the final result.

[0137] Thus, the data governance and audit log management system (100) based on an artificial intelligence model linkage interface according to an embodiment of the present invention unifies all data access between an artificial intelligence agent and a database through the system (100), thereby providing a technical effect of simultaneously achieving governance control before execution and audit log management after execution.

[0139] FIG. 3 is a flowchart showing the detailed flow of a data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention.

[0140] As shown in FIG. 3, the data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention may be configured to include (a) a data request reception step, (b) a policy judgment step before execution, (c) an execution control step, and (d) an audit log generation and storage step.

[0141] The problem to be solved by the data governance and audit log management method based on an artificial intelligence model linkage interface according to the embodiment of the present invention is to resolve data leakage, unauthorized access, excessive resource consumption, and difficulty in audit tracing that may occur when data requests generated by an artificial intelligence model or agent based on natural language-based queries are executed indiscriminately on an external data server.

[0142] In particular, the data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention provides a data governance and audit log management method that, considering the problem that it is difficult to predict in advance the execution results of data requests generated due to the characteristics of artificial intelligence models, performs a policy-based judgment at a stage prior to the actual execution of a data request and systematically records and manages the results.

[0143] In addition, the data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention can provide a technical effect that enables both pre-execution control and post-execution auditing in an AI-based data access environment by centrally processing data requests flowing in through the artificial intelligence model linkage interface and performing pre-execution policy judgment, execution control, and audit log recording in stages.

[0144] More specifically, as shown in FIG. 2, step (a) (S100) is a data request receiving step, which may be a step of receiving a natural language-based query from an artificial intelligence model or agent or a data request generated based on said natural language-based query.

[0145] At this stage, data requests may be received through a standardized integration interface (such as an MCP interface or a server) and may include the natural language query itself or an executable request transformed from the natural language query (such as SQL or API calls). The received data request may also include information to identify the requesting party, information regarding the time of the request, etc.

[0146] In this way, step (a) (S100) is a preprocessing step for introducing various types of data requests generated from different artificial intelligence models or agents into the same processing pipeline, and can provide a basis for consistently performing policy judgment and audit log generation in subsequent steps.

[0147] Through such step (a) (S100), the embodiment of the present invention can unify the input format of data requests even in various artificial intelligence model or agent environments, thereby preventing governance gaps caused by differences in the implementation of individual models.

[0148] In addition, by clearly recording the point of inflow of data requests, it can provide the effect of improving request traceability linked to subsequent audit logs.

[0149] And, step (b) (S200) is a pre-execution policy judgment step, which may be a step of determining whether to allow, restrict, or suspend the execution of a data request according to a pre-set data access policy before the received data request is executed on an external data server.

[0150] At this stage, the data access policy may be configured based on the authority or role of the requesting subject, the scope of the requested data, the type of request, whether sensitive information is included, etc.

[0151] Here, in step (b) (S200), the policy judgment is performed prior to the execution of the data request and may be a key step for controlling in advance whether the data request generated by the artificial intelligence model can be executed.

[0152] At this stage, in addition to simply passing or blocking the request, a decision may also be made to classify the request as pending if further review is required.

[0153] Based on such pre-execution policy judgment, the embodiment of the present invention provides a preemptive data governance effect that prevents the possibility of problems in advance, rather than a reactive approach where problems occur after a data request is actually transmitted to an external data server.

[0154] In particular, access control can be performed based on consistent standards for data requests automatically generated by AI agents, thereby simultaneously improving data security and operational stability.

[0155] And, as shown in FIG. 3, step (b) (S200) may include a step of predicting the risk or resource consumption cost associated with the execution of a data request based on the structural characteristics of the data request.

[0156] In a data governance and audit log management method according to an embodiment of the present invention, step (b) (S200) may include a step of predicting the risk or resource consumption cost associated with execution based on the structural characteristics of a received data request.

[0157] Here, the structural characteristics of the data request may include, for example, the query type, the range of the target data, the join structure, whether aggregate operations are included, and the complexity of the conditional clause; this structural information is utilized as input information to estimate in advance the potential for data exposure or system resource usage that may occur when the data request is actually executed.

[0158] This step is performed prior to the execution of the data request, and the prediction results are considered along with a pre-set data access policy and can be used as a basis for determining whether to execute.

[0159] That is, the method according to an embodiment of the present invention does not merely make a judgment based on whether the requesting entity has authority, but includes a process of analyzing the structure and nature of the data request itself to predict the execution result, thereby enabling the quantitative or qualitative evaluation of the potential impact of a data request generated by an artificial intelligence model or agent.

[0160] Accordingly, the data governance and audit log management method according to an embodiment of the present invention enables preemptive control considering risk and cost at a pre-execution stage, unlike conventional methods that recognize excessive resource consumption or exposure of sensitive information only after a data request is actually executed on an external data server.

[0161] In particular, data requests automatically generated by artificial intelligence models are often difficult to intuitively predict in advance, but in the embodiments of the present invention, such uncertainty can be effectively mitigated through prediction based on structural characteristics.

[0162] As a result, system stability, cost management efficiency, and data security can be improved simultaneously.

[0164] (c) Step (S300) is an execution control step, which may be a step of transmitting a data request to an external data server for execution, blocking transmission, or suspending processing based on the result of the judgment in Step (b).

[0165] In other words, data requests are transmitted to the external data server and executed only if the policy judgment result is “Allowed,” while if it is “Restricted,” the transmission of the data request is blocked or execution is prohibited. Additionally, if determined to be “Pending,” the data request may remain in a waiting state instead of being executed immediately.

[0166] This stage serves as a control phase that reflects the results of policy judgments into the execution flow of actual data requests, performing a technical role to ensure that data access policies do not remain merely declarative rules but are connected to actual execution control.

[0167] Accordingly, embodiments of the present invention can ensure the effectiveness of data access policies and structurally prevent the execution of data requests that may cause unauthorized data access or excessive resource consumption.

[0168] In addition, by providing room for further review regarding requests that are difficult to execute immediately through deferred processing, it is possible to simultaneously secure both the flexibility and security of data governance.

[0169] Additionally, if the processing of the data request is suspended in step (c) (S300) above, it is desirable that the data request be re-evaluated based on additional approval or policy changes.

[0170] That is, in the data governance and audit log management method according to the present embodiment, if a data request is determined to be pending as a result of execution control in step (c) (S300), the data request may be configured not to be executed immediately, but to be re-evaluated based on additional approval or policy changes.

[0171] In other words, the pending state does not simply block data requests, but functions as an intermediate state that suspends the final judgment on feasibility.

[0172] In addition, re-evaluation may be performed based on the approval results of additional authorized entities, such as security personnel and data governance managers, or after changes or supplements to the data access policy have been made.

[0173] These deferral and re-evaluation steps can provide a technical structure that prevents immediate execution for data requests generated by artificial intelligence models or agents when the risk or cost is high or the results of policy application are uncertain, while leaving open the possibility of future legal and safe execution.

[0174] Accordingly, in embodiments of the present invention, a flexible governance control mechanism can be implemented, going beyond the dichotomous control of “allow” and “block” for data requests.

[0175] In other words, by preventing the immediate execution of high-risk requests through a hold state, while enabling reasonable execution based on business needs or policy adjustments, it is possible to simultaneously ensure data security and business continuity.

[0176] As a result, the problem of excessive or insufficient control that may occur in an AI-based data access environment can be effectively mitigated.

[0177] Additionally, in step (c) (S300), the authenticated user's role and the access target (table, column) of the query to be executed may be compared, and unauthorized queries may be immediately blocked, and queries that pass access control may be transmitted to the DB for execution.

[0179] (d) Step (S400) is an audit log generation and storage step, which may be a step of generating and storing an audit log including at least one of requester information, request content, and judgment result for a data request.

[0180] Here, the audit log is intended to record the processing history from the receipt of a data request to the determination of whether to execute it, and may include the requesting party, the time of the request, the content of the request, and the policy judgment result. The generated audit logs are stored in an audit log repository and managed to enable retrospective retrieval.

[0181] This step can be performed regardless of whether a data request is executed, and can be configured to leave a consistent audit record for all allowed, restricted, or suspended data requests.

[0182] For example, in step (d) (S400), pre-execution information of the query (user ID, SQL, request time, etc.), execution results (success status, time taken, etc.), and query execution costs can be integrated and stored in the audit log store.

[0183] Through such step (d) (S400), the embodiment of the present invention can provide a basis for retrospectively tracking and analyzing data access behaviors of an artificial intelligence model or agent, thereby facilitating security incident analysis, compliance checks, internal audits, and dispute response, as well as providing technical effects that significantly enhance accountability and transparency in an artificial intelligence-based data access environment.

[0184] Additionally, step (d) (S400) may include a step of storing information indicating the reason for the decision on whether to execute the data request in the audit log.

[0185] That is, in the data governance and audit log management method according to an embodiment of the present invention, step (d) may include the step of storing information indicating the reason for determining whether to execute a data request in the audit log.

[0186] Information regarding the reason for judgment is information explaining why a data request was allowed, restricted, or withheld, and may include, for example, conditions of the applied data access policy, risk or cost prediction results, lack of authorization, or whether sensitive information is included.

[0187] The reasons for these judgments are recorded in the audit log along with the processing results of data requests, enabling the tracing of the basis for the judgment beyond a simple record of results.

[0188] In other words, in the embodiments of the present invention, transparency of the data request processing process can be ensured by recording not only the processing result of the data request but also the logical and policy reasons for the result.

[0189] With such a configuration, the embodiment of the present invention can provide a technical effect that can clearly explain the result of the judgment regarding a data request retrospectively.

[0190] This allows you to prove, based on objective records, why specific data requests were allowed or restricted during internal audits, compliance checks, security incident analysis, or disputes.

[0191] In particular, by ensuring explainability regarding the automated decision results of artificial intelligence models or agents, reliability and accountability in AI-based data access environments can be significantly enhanced.

[0193] In addition, the data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention may include, in step (a) (S100), a step of preprocessing a database query generated based on the data request to include a normalized representation or identification information for identifying its structural characteristics, and in step (d) (S400), a step of storing the normalized representation or identification information together in the audit log so as to enable repetitive access control and post-tracing for the same or substantially the same data request.

[0194] That is, in the data governance and audit log management method based on an artificial intelligence model linkage interface according to the present embodiment, the database query generated based on the data request in step (a) (S100) may include a preprocessing step in which the query is processed to include a normalized representation or identification information for identifying its structural characteristics.

[0195] Here, the normalized representation or identification information may include, for example, a standardized representation reflecting the structure of the query, an identifier based on a query tree (AST), or identification information expressing structural features as hash values.

[0196] This normalization or identification process is intended to ensure that substantially identical data requests are recognized identically, even if the surface representation of the query changes.

[0197] And, in step (d) (S400), the normalized representation or identification information is stored together in the audit log, which may include a step to identify whether the same or substantially identical data requests occur repeatedly.

[0198] That is, an embodiment of the present invention provides a method that enables identity determination based on structural characteristics without relying on string comparison of data requests.

[0199] With such a configuration, the embodiment of the present invention effectively identifies data requests and enables repetitive access control and post-tracing even when an artificial intelligence model or agent repeatedly generates the same data request with only different expressions.

[0200] As a result, system load caused by policy bypass attempts or excessive repetitive access can be prevented, and access pattern analysis and policy improvement based on audit logs become easier.

[0201] In particular, by structurally resolving the problem of repetitive requests that is prone to occur in AI-based automated query generation environments, the consistency and effectiveness of data governance can be significantly improved.

[0202] In addition, the data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention may further include a feedback and compliance response step.

[0203] The feedback and compliance response phase may be a phase of monitoring abnormal access patterns based on accumulated audit logs, and generating and outputting or reporting materials for data governance compliance using the data accumulated during monitoring.

[0205] FIG. 4 is a flowchart illustrating the step-by-step flow of a data governance and audit log management method based on an artificial intelligence model linkage interface according to an embodiment of the present invention.

[0206] As shown in FIG. 4, first, in step S100, a natural language-based query or a data request generated based on the natural language-based query is received from an artificial intelligence model or agent.

[0207] Next, in step S210, an authentication procedure for the requesting entity, such as token-based authentication, is performed, and in step S220, an access control decision is made regarding whether to allow the execution of the data request by comparing the authority of the requesting entity with the data access policy.

[0208] If the access control judgment result is approved, the process proceeds to step S300, and the query execution step for the database is performed.

[0209] On the other hand, if the access control judgment result is denied, the process does not proceed to the query execution stage, but moves to stage S420, and the fact that access to the data request was denied is recorded in the audit log.

[0210] Meanwhile, for approved data requests, after the query execution is performed in step S300, the audit log recording step is performed in step S410. At this time, metadata such as information prior to query execution, execution results, time taken for execution, and resource consumption costs are stored integrally in the audit log.

[0211] In the subsequent step S500, a feedback phase is conducted to monitor abnormal access patterns based on the stored audit logs and perform compliance responses or reporting if necessary.

[0212] As such, the method according to the present invention provides a method for controlling data access by an artificial intelligence model or agent in advance and tracking and managing it subsequently by systematically connecting the entire process from the reception of a data request to authentication, access control, execution, audit log recording, and subsequent response.

[0214] In addition, another embodiment of the present invention may be characterized by a computer program stored on a storage medium to execute a data governance and audit log management method based on an artificial intelligence model linkage interface on a computer.

[0215] In addition, the program applied to the data governance and audit log management method based on an artificial intelligence model linkage interface according to one embodiment of the present invention may be implemented as computer-readable code on a computer-readable recording medium. The codes and code segments implementing the above program can be easily inferred by a computer programmer in the field.

[0216] Here, a computer-readable recording medium may include any type of recording device in which data that can be read by a computer system is stored. Examples of computer-readable recording media may include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical disk, etc. Additionally, computer-readable recording media may be distributed across networked computer systems and may be written and executed as computer-readable code in a distributed manner.

[0218] Below, more specific embodiments that can be applied to the data governance and audit log management system (100) and method based on an artificial intelligence model linkage interface according to an embodiment of the present invention will be described in detail.

[0220] 1. Immutable metadata encompassing authentication, authorization, and query contexts ( AuthEnvelope )

[0221] A data governance and audit log management system (100) and method based on an artificial intelligence model linkage interface according to one embodiment of the present invention can generate an authentication, authorization, and query context package (e.g., AuthEnvelope, AuditEnvelope, etc.) to capture “who (principal), what (query), what context and under what policy the request is made” as immutable metadata at a stage prior to a data request being executed on an external data server.

[0222] At this time, the AuthEnvelope may include identification information of the requesting principal (e.g., user_id, agent_id, role), request context (e.g., natural language input, generated SQL / request), authentication token and session information, policy information used for evaluation (or policy snapshot identifier), and through this metadata, a deterministic structure can be provided that makes the same judgment reproducible for the same input.

[0223] In particular, by including a hash (e.g., sql_norm_ast_hash) generated by normalizing SQL (AST-based), a basis for determining query identity that is not affected by superficial changes such as blanks or aliases is established, and by including policy_snapshot which fixes the role-based access control and domain rules at the time of evaluation as a snapshot, the judgment can be reproduced retrospectively using “rules at that time”.

[0224] In addition, by standardizing and recording feature sets (e.g., risk_features) for predicting risks and costs before execution, policy judgments, prediction model inputs, and audit records can be linked, and access control information (e.g., column_level_access) that subdivides role-based access control at the column / cell level can be included to deterministically implement masking and blocking of sensitive information.

[0226] 2. Decision and Reason Code Centralized audit log schema ( AuditLog ) and full-cycle transparency

[0227] In one embodiment, the audit log management unit (150) (or step (d)) may apply an audit log schema that goes beyond simply recording “executed / failed” and stores the prediction result + reason code together.

[0228] For example, if the policy judgment result is a denial, reason codes such as “ROLE_MISMATCH” or “PII_ACCESS_DENIED” can be recorded; if it is a hold, reason codes such as “COST_CAP_EXCEEDED” or “RISK_HIGH” and necessary additional approval entities (e.g., IT_AUDIT, Data_Governance) can be recorded.

[0229] Thus, in one embodiment of the present invention, transparency is achieved throughout the entire lifecycle, extending beyond the “recording of execution results” to include the “basis for prediction and policy judgment prior to execution,” thereby providing an auditable structure advantageous for regulatory compliance, prevention of post-dispute resolution, and improvement of repetitive decision-making.

[0231] 3. Pre-pipeline and steps prior to the query reception and authentication stage (a) Specific examples of preprocessing

[0232] In one embodiment of the present invention, as a preprocessing step that may be included in step (a) (S100), a structured pipeline from natural language input to data request generation and pre-execution determination may be further included.

[0233] For example, the sequence may include (1) input normalization and standardization, (2) intent classification (domain / purpose), (3) schema linking (table / column mapping), (4) NL→SQL candidate generation, (5) SQL normalization (AST construction and canonical hash), (6) risk / cost / sensitivity feature extraction, (7) policy interpretation and snapshot generation (RBAC+domain rule), (8) risk / cost prediction before execution (ML inference), (9) approval / hold / rejection decision orchestration and reason code assignment, (10) Envelope and audit log storage, (11) DB execution only upon approval.

[0234] Here, the schema linking step can map natural language entities to actual tables / columns by referencing schema / dictionary tables and domain ontology, etc., and the SQL normalization step enables query identity determination, cache key generation, and duplicate detection through AST-based normalization and hashing.

[0235] Additionally, the feature extraction step may include syntax features such as the number of joins, number of aggregations, subquery depth, and where selectivity, sensitivity features such as PII / medical / financial, EXPLAIN, or statistics-based cost estimation (e.g., estimated_cost_usd).

[0237] 4. Policy Snapshot Basic policy interpretation and decision-making of orchestration Examples

[0238] In one embodiment, policy resolution can generate a snapshot by statically combining role-based access control (RBAC: role → table / column / query type authority) and domain-specific rules (e.g., PHI / PII masking, re-identification risk, cost caps, operating time limits, etc.) and include it in subsequent decision-making.

[0239] In addition, ML-based risk prediction (classification) and cost prediction (regression) can be performed before execution, and branch to hold or block if a threshold is exceeded. Decision orchestration can be implemented by applying decision trees (decision rules) in the order of “policy priority → risk score → cost cap”, for example, and assigning reason codes to the final result.

[0240] Specifically, the procedure for evaluating a request may be performed in the following order: (1) authorization check (RBAC), (2) policy snapshot loading, (3) application of sensitivity rules, (4) cost prediction and comparison of per-user cost limits (hold if exceeded), (5) risk prediction and comparison of thresholds (hold if exceeded), (6) application of column-level masking / rewriting, and (7) final approval.

[0242] 5. column / Cell-level access control and automatic rewrite( masking / Scroll down) Examples

[0243] In one embodiment of the present invention, RBAC can be subdivided beyond the table level to the column / cell level, and masking or blocking of sensitive information by domain can be deterministically implemented.

[0244] Additionally, if a data request (e.g., SQL) includes sensitive columns, automatic rewrite can be performed according to the policy; for instance, this can replace a column like customer_ssn with mask(customer_ssn) or reduce the scope by forcibly inserting a WHERE condition for large-volume queries that lack a specific period filter.

[0245] This rewriting goes beyond the “allow / block” dichotomy and serves as an example of transforming requests into a secure form by directly reflecting policies at the execution layer.

[0247] 6. Pre-execution prediction model and hybrid of the learning structure (inference is real-time, learning is isolated) Examples

[0248] In one embodiment of the present invention, “risk or resource consumption cost prediction” can be implemented through ML inference performed in a pre-execution step and may include a Risk Classifier and a Cost Regressor.

[0249] The risk classifier takes the number of joins, sensitivity flags, query types, user roles, past rejection history, outlier features, etc. as input, and can apply, for example, a tree-based model or an AST embedding-based model.

[0250] The cost regressionr takes estimated scan / IO, historical execution time / cost, DB statistics, etc., as input and may include engine-specific EXPLAIN parameter correction, and may include online adaptation that dynamically readjusts thresholds by user / role / domain.

[0251] In addition, in one embodiment of the present invention, the learning structure can be configured as a hybrid for operational stability and security.

[0252] For example, real-time inference is performed on an intermediary server (e.g., an inference plane executed by an MCP server), while the core model (risk / cost model) is batch trained on a separate, isolated training server and version-distributed via a model registry, allowing only lightweight adaptive parameters to be periodically updated on the inference plane.

[0253] At this time, the exposure of sensitive data can be reduced by loading PII / PHI columns into the Feature Store with masking / tokenization and having the model train / infer based on sensitivity “flags,” and decision reproducibility can be guaranteed through policy snapshot IDs.

[0255] 7. Implementation details (example) of Claim 9 (Normalization / Identification Information) and "Repetitive Access Control / Post-Trace"

[0256] In one embodiment of the present invention, the normalized expression or identification information may be implemented by converting SQL into an AST using an SQL parser, removing literals / spaces / aliases, and standardizing the order of operators and syntax to generate a hash value.

[0257] The generated hash can be used for duplicate query detection, policy cache keys, and ensuring audit reproducibility, and the corresponding normalized identifier is stored in the audit log to enable repetitive access control and post-hoc tracing for identical or substantially identical requests.

[0258] In addition, if results are returned from the DB in the post-execution phase, the system can be configured to record both the predicted cost / risk and the actual execution cost (or execution time) to enable a “predicted vs. actual” comparison, which can be used as input for policy threshold adjustment and online adaptive parameter correction.

[0260] Through the data governance and audit log management system (100) and method based on an artificial intelligence model linkage interface according to one embodiment of the present invention described above, the following effects can be obtained due to technical differentiation.

[0261] 1) With a formalized pipeline and normalized AST hash in the pre-execution stage, query identity and reproducibility, as well as caching and duplicate detection, can be systematically resolved.

[0262] 2) By incorporating Policy Snapshots into decision-making, decisions can be reproduced even with changes in rules over time. (Audit-friendly)

[0263] 3) Automatic blocking, holding, and approval orchestration can be implemented by inferring risk / cost prediction models in real-time on the Front (MCP).

[0264] 4) Domain rules can be directly applied to the execution layer through column / cell level RBAC and automatic rewrite (masking / filter).

[0265] 5) With a hybrid learning structure, sensitive data and DB load can be separated from the learning environment, inference can be performed in real time, and learning can be isolated to enhance operational stability and security.

[0267] Although various preferred embodiments of the present invention have been described above with some examples, the descriptions of various embodiments described in the "Specific details for carrying out the invention" section are merely illustrative, and those skilled in the art to which the present invention pertains will understand that the present invention can be modified in various ways or equivalent embodiments can be carried out based on the above description.

[0268] In addition, since the present invention can be implemented in various other forms, the present invention is not limited by the description above. The above description is provided merely to make the disclosure of the present invention complete and to fully inform those skilled in the art of the scope of the present invention, and it should be understood that the present invention is defined only by each claim of the claims. Explanation of the symbols

[0269] 100: Data Governance and Audit Log Management System 110: Standardized interlocking interface section 111 : Interface provision module 112: Request reception and validation module 113: Normalization and Structuring Module 114 : Request Context Creation Module 130 : Governance Control Unit 131: Policy Storage and Loading Module 132 : Requesting Subject Authentication Module 133 : Risk and Cost Prediction Module 134 : Policy Evaluation Module 135: Execution Decision Orchestration Module 136 : Pending Processing Linkage Module 150 : Audit Log Management Department 151: Schema Definition and Creation Module 152 : Log storage module 153 : Log Integrity Protection Module 154 : Log Inquiry and Reporting Module

Claims

Claim 1 A standardized interoperability interface unit for receiving a natural language-based query or a data request generated based on said natural language-based query from an artificial intelligence model or agent; and a governance control unit that identifies the requesting entity by verifying credentials included in said data request, and controls the execution of said data request to allow, restrict, or suspend according to a pre-set data access policy before said data request is executed on an external data server while blocking the user or agent from directly accessing the external data server. A data governance and audit log management system based on an artificial intelligence model linkage interface, comprising an audit log management unit that generates and stores an audit log including at least one of information on the requesting entity, the content of the request, and the judgment result of the governance control unit regarding the data request, wherein the governance control unit predicts the risk of data exposure or resource consumption costs associated with the execution of the data request based on the structural characteristics of the data request, and determines whether to execute the data request by considering the prediction result together with the data access policy; wherein the audit log management unit enables the subsequent reproduction of the judgment result regarding the data request by including and storing policy information or policy identification information regarding the data access policy applied to the data request in the audit log, and, when the data request is executed, includes and stores the execution result of the data request and the time or execution cost incurred in the execution in the audit log. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 (a) receiving a natural language-based query or a data request generated based on said natural language-based query from an artificial intelligence model or agent; (b) verifying credentials included in said data request to identify the requesting entity, and determining whether to allow, restrict, or suspend the execution of said data request according to a pre-set data access policy before said data request is executed on an external data server while blocking the user or agent from directly accessing the external data server; (c) transmitting said data request to said external data server for execution, blocking transmission, or suspending processing according to said determination result; A method for managing data governance and audit logs based on an artificial intelligence model linkage interface, comprising: (d) a step of generating and storing an audit log including at least one of requester information, request content, judgment result, and execution result for the data request, wherein step (b) includes a step of predicting the risk or resource consumption cost associated with the execution of the data request based on the structural characteristics of the data request and a step of determining whether to execute the data request by considering the prediction result together with the data access policy; and step (d) a step of enabling subsequent reproduction of the judgment result for the data request by including and storing policy information or policy identification information regarding the data access policy applied to the data request in the audit log, and, if the data request is executed, including and storing the execution result of the data request and the time or execution cost incurred in the execution in the audit log. Claim 6 delete Claim 7 delete Claim 8 A method for managing data governance and audit logs based on an artificial intelligence model integration interface, characterized in that, in claim 5, if the processing of the data request is suspended in step (c), the data request is re-evaluated according to additional approval or policy change. Claim 9 A method for data governance and audit log management based on an artificial intelligence model integration interface according to claim 5, characterized in that, in step (a), the database query generated based on the data request is preprocessed to include a normalized representation or identification information for identifying its structural characteristics, and in step (d), the normalized representation or identification information is stored together in the audit log to enable repetitive access control and post-tracing for the same or substantially the same data request. Claim 10 A computer program characterized by being stored on a storage medium to execute a data governance and audit log management method based on an artificial intelligence model linkage interface of any one of claims 5, 8, and 9 on a computer.

Images