An autonomous access control method, system, device and medium for agent data

Abstract

The application belongs to the technical field of data access, and discloses an agent data autonomous access control method, system, equipment and medium for a trusted data space. The control method comprises the following steps: first, extracting the data access constraint of the trusted data space digital contract, generating a constraint description file through compilation and signature, and issuing the constraint description file to an agent operating environment; when the agent initiates a data operation, automatically generating an intention declaration, performing multi-dimensional compliance verification and calculating a compliance degree score based on the constraint file, and executing a hierarchical response accordingly; before the agent outputs, scanning the output content in combination with the constraint file and the original data, automatically correcting the rule violation, and directly outputting the rule compliance. The scheme converts the contract rules into executable constraints of the agent, realizes real-time pre-event and in-process fine management and control of the autonomous operation of the agent, does not need frequent manual approval, synchronously intercepts the rule violation output, reduces the data leakage risk from the source, and balances the automation efficiency of the agent and the compliance of cross-organization data use.

Description

Technical Field

[0001] This application belongs to the field of data access technology, and specifically relates to a method, system, device and medium for intelligent agent autonomous data access control in a trusted data space. Background Technology

[0002] With the rapid development of AI Agent (Artificial Intelligence Agent) technology, intelligent agents have evolved from simple dialogue tools into automated systems with autonomous perception, decision-making, and execution capabilities. In enterprise applications, AI Agents can autonomously plan tasks, invoke tool interfaces, access multi-source data, and generate analysis results, significantly improving business efficiency. However, when AI Agents need to access and use data across organizations in a trusted data space, traditional data usage control mechanisms face entirely new challenges. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a method, system, device, and medium for autonomous access control of intelligent agents in a trusted data space. This solution enables real-time, pre-emptive, and in-process refined control of autonomous agent operations, reducing the risk of data leakage at the source while balancing the automation efficiency of intelligent agents with the compliance of cross-organizational data usage.

[0004] The first aspect of this disclosure proposes a method for autonomous access control of intelligent agents in a trusted data space, the control method comprising: Obtain data access constraints from the digital contracts in the trusted data space, compile and sign the data access constraints to generate a constraint description file, and send it to the intelligent agent's runtime environment; When the intelligent agent performs data operations on the trusted data space, it generates an intent statement corresponding to the data operation, performs multi-dimensional compliance verification on the intent statement based on the constraint description file to generate a compliance score, and responds to the data operation in a graded manner according to the compliance score. Before the agent completes data operations and outputs data, a compliance scan is performed on the output data based on the constraint description file and the original data; if the output data violates the rules, the violation content in the output data is corrected; if the output data is compliant, the output data is output.

[0005] According to a preferred embodiment of this disclosure, the control method further includes: Obtain the intent statement, the compliance score, the response type of the graded response, the data operation, the output data, the operation time of the data operation, and the resource consumption; The intent statement, compliance score, response type of graded response, data operation, output data, operation time and resource consumption of the data operation are linked in chronological order, and the hash value is calculated according to the event node to obtain the behavior trajectory chain; The digital contract is associated with the behavior trajectory chain, and a compliance proof report corresponding to the digital contract is generated based on the operation status of the data operation, the actual execution status of each data access constraint in the constraint description file, and the scanning results of the compliance scan.

[0006] According to a preferred embodiment of this disclosure, the step of obtaining data access constraints in the digital contract of the trusted data space, compiling and signing the data access constraints to generate a constraint description file, and then sending it to the intelligent agent's runtime environment includes: Extract the contract target, operation blacklist / whitelist, time constraint, frequency constraint, environmental constraint, and obligation clause constraint from the digital contracts in the trusted data space; Based on the contract subject matter, operation blacklist / whitelist, time constraints, frequency constraints, environmental constraints, and obligation clause constraints, a capability envelope containing a unique envelope identifier, authorized data products and fields, operation whitelist, single query limit, output de-identification rules, cache expiration time, and validity period is compiled and generated. The capability envelope is signed using the data provider's key to generate a constraint description file, which is then sent to the agent's runtime environment.

[0007] According to a preferred embodiment of this disclosure, when the intelligent agent performs data operations on the trusted data space, it generates an intent statement corresponding to the data operation, performs multi-dimensional compliance verification on the intent statement based on the constraint description file to generate a compliance score, and performs a graded response to the data operation based on the compliance score, including: When the intelligent agent performs data operations on the trusted data space, it generates an intent declaration that includes an operation identifier, accessed data source, operation type, query parameters, business purpose, and upstream call chain. Based on the constraint description file, the intent declaration is subjected to compliance checks for data range, operation type, query parameters, and cache expiration time. The compliance score is calculated by weighted average of the compliance check results. When the compliance score determines that the operation is compliant, the data operation is allowed; when the compliance score determines that the operation is slightly out of bounds, the operation level of the data operation is downgraded; when the compliance score determines that the operation is in violation, the data operation is stopped and an alarm is issued. When the intent statement exceeds the scope of the constraint description file, manual review is required to determine whether to extend the constraint description file or reject the data operation.

[0008] According to a preferred embodiment of this disclosure, downgrading the operation level of the data operation includes: Convert the record-by-record detailed query in the data operation into an aggregated statistical query; The data-sensitive fields of the output data corresponding to the data operations are automatically de-identified and masked. The number of data entries returned in a single query of the data operation is compressed to within the limited threshold of the digital contract.

[0009] According to a preferred embodiment of this disclosure, the step of performing a compliance scan on the output data based on the constraint description file and the original data; when the output data is found to be non-compliant, a correction operation is performed on the non-compliant content in the output data, including: Calculate the text similarity and structured field matching degree between the original data and the output data; A data leakage risk value is generated by weighted fusion of the text similarity and the structured field matching degree; Determine whether the output data violates the constraint description file and the data leakage risk value; When the output data is invalid, the data in the output data is replaced with an aggregated statistical description, or generalized, or rounded.

[0010] According to a preferred embodiment of this disclosure, the control method further includes: Add a globally unique ID to the constraint description file, and associate the globally unique ID with the digital contract and the data operation.

[0011] To address the aforementioned technical problems, a second aspect of this disclosure proposes an intelligent agent data autonomous access control system for trusted data spaces, the control system comprising: The constraint compilation module is used to obtain data access constraints in the digital contracts of the trusted data space, compile and sign the data access constraints to generate a constraint description file and send it to the intelligent agent's runtime environment. The operation response module is used to generate an intent statement corresponding to the data operation when the agent performs a data operation on the trusted data space, perform multi-dimensional compliance verification on the intent statement based on the constraint description file to generate a compliance score, and perform a graded response to the data operation according to the compliance score. The data output module is used to perform a compliance scan on the output data based on the constraint description file and the original data before the intelligent agent completes the data operation and outputs the data; when the output data is non-compliant, it corrects the non-compliant content in the output data; when the output data is compliant, it outputs the output data.

[0012] To address the aforementioned technical problems, a third aspect of this disclosure provides an electronic device, comprising: Processor; and A memory storing computer-executable instructions, which, when executed, cause the processor to perform the control method described in any of the above embodiments.

[0013] To address the aforementioned technical problems, a fourth aspect of this disclosure provides a computer storage medium that stores one or more programs, which, when executed by a processor, implement the control method described in any of the above embodiments.

[0014] Compared with existing technologies, this application has the following advantages: This solution first extracts the data access constraints of digital contracts in the trusted data space, compiles and signs them to generate constraint description files, and then distributes them to the intelligent agent's runtime environment; when the intelligent agent initiates a data operation, an intent declaration is automatically generated, and multi-dimensional compliance verification is performed based on the constraint file to calculate the compliance score, thereby executing a tiered response; before outputting, the intelligent agent scans the output content in conjunction with the constraint file and the original data, and automatically corrects any violations or directly outputs compliant data. This solution transforms contract rules into executable constraints for the intelligent agent, enabling real-time, pre- and in-process refined control of the intelligent agent's autonomous operation, eliminating the need for frequent manual approvals, and simultaneously intercepting illegal outputs, thereby reducing the risk of data leakage from the source and balancing the automation efficiency of the intelligent agent with the compliance of cross-organizational data use.

[0015] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1A schematic flowchart of an intelligent agent's autonomous access control method for trusted data space, according to an embodiment of the present disclosure, is shown. Figure 2 A second schematic flowchart of an agent-based autonomous access control method for trusted data space, according to an embodiment of the present disclosure, is shown. Figure 3 A schematic flowchart of a method for autonomous access control of intelligent agent data in a trusted data space, according to an embodiment of the present disclosure, is shown in part three. Figure 4 A schematic flowchart of a method for autonomous access control of intelligent agent data in a trusted data space, according to an embodiment of the present disclosure, is shown in Part IV. Figure 5 The fifth illustration shows a flowchart of an agent-based autonomous access control method for trusted data spaces according to an embodiment of the present disclosure; Figure 6 A schematic diagram of an intelligent agent data autonomous access control system structure for a trusted data space is shown according to an embodiment of the present disclosure; Figure 7 A schematic diagram of an electronic device structure according to an embodiment of the present disclosure is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The same reference numerals in the accompanying drawings denote the same or similar elements, components, or parts, and therefore repeated descriptions of the same or similar elements, components, or parts may be omitted below. It should also be understood that although qualifiers such as first, second, third, etc., indicating numbers may be used herein to describe various devices, elements, components, or parts, these devices, elements, components, or parts should not be limited by these qualifiers. That is, these qualifiers are only used to distinguish one from another. For example, a first device may also be referred to as a second device, without departing from the essence of the technical solution of this disclosure. Furthermore, the terms "and / or" and "and / or" refer to all combinations including the first or more of the listed items.

[0020] Please see Figure 1 , Figure 1 This is one of the flowcharts of an intelligent agent's autonomous access control method for trusted data spaces provided in this disclosure, such as... Figure 1As shown, the control methods include: S11. Obtain the data access constraints in the digital contract of the trusted data space, compile and sign the data access constraints to generate a constraint description file and send it to the intelligent agent's runtime environment.

[0021] In this embodiment, data access constraints within the trusted data space digital contract are extracted, compiled into a machine-readable constraint description file, digitally signed, and then distributed to the intelligent agent's runtime environment. This achieves the automatic conversion of digital contract business rules into enforceable constraints at the intelligent agent's runtime. The digital signature ensures the constraint file is tamper-proof, and the targeted distribution of the constraint file can be bound to intelligent agent tasks to form independent permission boundaries. This connects the contract and intelligent agent permission control systems, providing a unified and reliable basis for subsequent compliance verification.

[0022] In this embodiment, a globally unique ID is added to the constraint description file, and this globally unique ID is associated with the digital contract and data operations. The globally unique ID can be associated with the task ID and contract ID; a one-to-one binding relationship is established between the contract, constraint file, and intelligent agent operation, enabling precise traceability of the entire chain of behavior. Relying on the unique identifier, a complete chain of evidence is linked, which facilitates the subsequent verification of whether the intelligent agent's data usage behavior matches the contract agreement, forming verifiable compliance audit evidence.

[0023] S12. When the agent performs data operations on the trusted data space, it generates an intent statement for the corresponding data operation, performs multi-dimensional compliance verification on the intent statement based on the constraint description file to generate a compliance score, and responds to the data operation in a graded manner according to the compliance score.

[0024] In this embodiment, when an agent initiates a data operation, a corresponding intent statement is generated. Multi-dimensional compliance verification is carried out based on the constraint description file, and a compliance score is calculated. The data operation is responded to in a graded manner according to the score. It can identify various autonomous data access behaviors of the agent in real time, accurately distinguish different degrees of violations through quantitative scoring, and handle them differently. Compliant operations are automatically allowed to ensure the efficiency of the agent's operation, and abnormal operations are controlled in a timely manner. This solves the problem of manual approval dragging down automation capabilities and achieves dynamic real-time compliance interception without sacrificing the autonomy of the agent.

[0025] S13. Before the agent completes data operations and outputs data, it performs a compliance scan on the output data based on the constraint description file and the original data. When the output data is non-compliant, it performs correction operations on the non-compliant content in the output data. When the output data is compliant, it outputs the output data.

[0026] In this embodiment, before the agent outputs data, an output compliance scan is performed by combining the constraint description file and the original data. Violations are identified and automatically corrected, while compliant content is directly output. This extends data security control to the output stage, proactively intercepting the risk of leakage of original details and identifiable sensitive information, eliminating the need for post-event tracing, blocking data leakage from the terminal, and automatically correcting to avoid task interruption, thus balancing compliance constraints with the business continuity of agent analysis output.

[0027] In this embodiment, the solution first extracts the data access constraints of the trusted data space digital contract, compiles and signs them to generate a constraint description file, and then distributes it to the intelligent agent's runtime environment. When the intelligent agent initiates a data operation, an intent declaration is automatically generated. Based on the constraint file, multi-dimensional compliance verification is performed and a compliance score is calculated, and a tiered response is executed accordingly. Before outputting, the intelligent agent scans the output content using the constraint file and the original data. If there are violations, the output is automatically corrected; otherwise, it is directly output in compliance with the rules. This solution transforms contract rules into executable constraints for the intelligent agent, enabling real-time, pre- and in-process refined control of the intelligent agent's autonomous operation. It eliminates the need for frequent manual approvals, simultaneously intercepts illegal outputs, reduces the risk of data leakage from the source, and balances the automation efficiency of the intelligent agent with the compliance of cross-organizational data use.

[0028] Please see Figure 2 , Figure 2 This is a schematic diagram of the second part of a method for intelligent agent autonomous access control of trusted data space provided in this disclosure. Figure 2 As shown, the control methods include: S21. Obtain the intent statement, compliance score, response type of graded response, data operation, output data, operation time of data operation, and resource consumption.

[0029] In this embodiment, a unified collection of multi-dimensional operation information, including intent declaration, compliance score, graded response type, data operation content, output data, operation time, and resource consumption, is used to completely retain the details of the entire process of intelligent agent data access. This allows for the construction of complete and detailed operation records, supporting the subsequent generation of tamper-proof trajectory chains, issuance of compliance certificates, and multi-dimensional behavior tracing. It provides complete original evidence for data compliance auditing, abnormal behavior location, and resource usage statistics.

[0030] S22. Link the intent statement, compliance score, response type of graded response, data operation, output data, operation time of data operation and resource consumption in chronological order, and calculate the hash value of the event node to obtain the behavior trajectory chain.

[0031] In this embodiment, the collected intelligent agent's entire process operation information is sorted by time as event nodes, and then linked sequentially and hash values ​​are calculated for each node to generate a behavior trajectory chain. The chain hashing feature ensures that the entire operation record is tamper-proof, completely solidifying all behavioral details of the intelligent agent from initiating access to outputting results, forming credible and irrefutable audit evidence, and supporting the data provider to quickly verify whether the intelligent agent's data usage behavior complies with the requirements of the digital contract.

[0032] In this embodiment, during the Agent's task execution, the system automatically records all data operation events, forming a complete "behavioral trajectory chain." Each event node in the trajectory chain includes: the complete content of the intent declaration, the comparison result and response type (allow / downgrade / block) of the policy interception layer, the actual data operation performed and its return result summary (such as the number of returned records, aggregation dimension, excluding the original data), operation time, and resource consumption. The event nodes are linked in chronological order, and chain hashing ensures the integrity and immutability of the trajectory.

[0033] In this embodiment, the chain hash calculation formula for the behavior trajectory chain is as follows:

[0034] in, The hash value of the nth event node. The hash value of the previous event node (the hash value of the first event node). (The combined hash of the data operation task ID and the capability envelope (i.e., the constraint description file)). This is the identifier for this operation. For timestamps, This is a summary of the intended content. For the policy-based interception layer's response decision (allow / downgrade / block). This is a summary of the actual data operation results, where || represents the field concatenation operation. The SHA-256 algorithm is used. Any tampering with an event node will cause the hash values ​​of all subsequent nodes to be inconsistent, thus being detected by integrity checks.

[0035] S23. Associate the digital contract with the behavior trajectory chain, and generate a compliance proof report for the corresponding digital contract based on the operation status of the data operation, the actual execution status of each data access constraint in the constraint description file, and the scanning results of the compliance scan.

[0036] In this embodiment, a binding association is established between the digital contract and the hash-fixed behavior trajectory chain. By combining all data operation records of the intelligent agent, the implementation status of various access constraints in the constraint description file, and the output compliance scan results, a compliance proof report matching the digital contract is automatically generated. This can intuitively present whether the intelligent agent's data use throughout the entire process complies with the contract agreement, forming a standardized, deliverable, and verifiable compliance certificate, eliminating information asymmetry between data supply and demand parties, and improving trust in cross-organizational data collaboration.

[0037] In this embodiment, upon completion of the task, the system associates the behavioral trajectory chain with the digital contract that triggered the task, automatically generating a "compliance verification report." The report includes: the contract identifier and capability envelope identifier; the total number of queries during task execution, the distribution of each operation type, the number of blocked operations and their reasons; the actual execution status of each constraint in the capability envelope (e.g., whether the query frequency is within the limit, whether the cache is cleared within the specified time); and the final compliance scan results. The compliance verification report is signed by the Agent runtime environment and submitted to the data provider's connector, allowing the data provider to verify that its data has indeed been used in accordance with the contract.

[0038] Furthermore, this solution can query behavioral trajectory chains and compliance certification reports as needed, and determine whether data usage is reasonable based on these chains and reports. Specifically, it allows for the tracing of abnormal behavior; when a data provider questions data usage, they can request to view the complete behavioral trajectory chain. The tracing process supports filtering and searching by operation type, time range, data fields, and other dimensions. For operations that have been downgraded, the trajectory chain records the original intent statement before downgrade and the actual execution content after downgrade, enabling the data provider to determine whether the downgrade was reasonable.

[0039] Please see Figure 3 , Figure 3 This is the third flowchart of a method for autonomous access control of intelligent agent data in a trusted data space, as provided in this disclosure. Figure 3 As shown, the control methods include: S31. Extract contract targets, operation blacklists and whitelists, time constraints, frequency constraints, environmental constraints, and obligation clause constraints from digital contracts in the trusted data space.

[0040] In this embodiment, various constraints, including contract targets, operation blacklists and whitelists, time, frequency, environment, and obligation clauses, are extracted from the digital contract in the trusted data space. This comprehensively covers all dimensions of data access control rules and fully collects all usage restrictions stipulated in the contract. This provides a complete and comprehensive original rule basis for the subsequent compilation and generation of constraint description files with refined control capabilities, ensuring that the intelligent agent's data access control can fully implement all compliance requirements of the digital contract.

[0041] In this embodiment, when the AI ​​Agent is authorized to execute a task involving trusted data space data, the system first extracts the contract policy from the associated digital contract. The extraction of the contract policy is completed through the contract management interface of the trusted data space connector. The extracted content includes contract target information (data product ID involved, data field range), operational behavior constraints (allowed and prohibited operation types), constraints (time constraints, purpose constraints, subject constraints, security environment constraints, etc.), and obligation clauses (obligation to record audit logs, obligation to delete after use, etc.).

[0042] S32. Based on the contract target, operation blacklist / whitelist, time constraints, frequency constraints, environmental constraints, and obligation clause constraints, compile and generate capability envelopes containing a unique envelope identifier, authorized data products and fields, operation whitelist, single query limit, output desensitization rules, cache expiration time, and validity period.

[0043] In this embodiment, the multi-dimensional constraint content extracted from the digital contract is compiled into a capability envelope carrying multiple restrictions such as unique identifier, data access permission, query, output, caching, and timeliness. This transforms the contract text rules into a standardized constraint carrier that can be automatically recognized and verified by machines, realizing the automatic conversion of contract terms into runtime management rules for intelligent agents. This provides a unified, complete, and refined judgment benchmark for subsequent compliance verification of intent declarations and review of output content.

[0044] In this embodiment, the system compiles the extracted contract policy into a structured "Capability Envelope," which is a machine-executable constraint description file oriented towards the agent runtime. The data structure of the Capability Envelope includes at least the following fields: Envelope Identifier: A globally unique capability envelope ID, associated with the task ID and contract ID; Accessible Data Product List: The set of data product IDs that the Agent is allowed to access in this task and the range of fields that are allowed to be accessed; Operation Type Whitelist: The set of data operation types that the Agent is allowed to execute, including read-only queries, aggregation statistics, feature extraction, model training, etc. Operation types not in the whitelist are prohibited by default; Single Query Constraints: Including the maximum number of records returned in a single query, the minimum aggregation granularity (e.g., no less than 10 records should be returned in the aggregation result), and the upper limit of query frequency (e.g., no more than N queries per minute); Output Constraints: Including prohibiting the inclusion of original records in the output, prohibiting the output of identifiable personal information, and the maximum precision of the output results (e.g., numerical results should be retained to two decimal places); Caching Constraints: The maximum caching time for intermediate results (e.g., caching should not exceed T minutes), and prohibiting the writing of query results to the Agent's long-term memory storage; Envelope Validity Period: A time window that is the same as or shorter than the contract validity period.

[0045] S33. Sign the capability envelope based on the data provider's key to generate a constraint description file, and then send the constraint description file to the agent's operating environment.

[0046] In this embodiment, the compiled capability envelope is digitally signed using the data provider's key to obtain a constraint description file, which is then distributed to the agent's runtime environment. The digital signature can prevent various access constraints within the capability envelope from being tampered with, ensuring the authenticity and reliability of the control rules. At the same time, the standardized constraint carrier is distributed and bound to the corresponding agent, providing an immutable and authoritative basis for real-time compliance verification of all subsequent data operations of the agent.

[0047] In this embodiment, the compiled capability envelope is signed by the data provider's connector using its digital signature key, ensuring that the envelope content is immutable. The signed capability envelope is then injected into the Agent's runtime environment through a secure channel. The injection process of the capability envelope is transparent to the Agent's business logic; that is, the Agent's task planning and inference logic do not need to be aware of the envelope's existence—the envelope only operates at the interaction layer between the Agent and the data source.

[0048] Please see Figure 4 , Figure 4 This is the fourth flowchart of a method for autonomous access control of intelligent agent data in a trusted data space, as provided in this disclosure. Figure 4 As shown, the control methods include: S41. When the agent performs data operations on the trusted data space, it generates an intent declaration that includes the operation identifier, access data source, operation type, query parameters, business purpose, and upstream call chain.

[0049] In this embodiment, when an agent initiates a data operation, an intent declaration is automatically generated, which includes multi-dimensional information such as operation identifier, access data source, operation type, query parameters, business purpose, and upstream call chain. This fully restores the entire context of the agent's data access, providing complete and fine-grained judgment material for subsequent multi-dimensional compliance verification and violation location. It accurately restores the agent's autonomous chain access behavior and supports refined hierarchical management.

[0050] In this embodiment, during the Agent's task execution, whenever the Agent needs to initiate a data operation (such as query, aggregation, export, etc.), the Agent runtime framework (such as the tool call layer based on the MCP protocol) automatically generates a structured "Intent Declaration". The generation process of the Intent Declaration is transparent to the Agent's business logic and is automatically completed by the tool call interceptor of the Agent runtime framework. The data structure of the Intent Declaration includes at least the following fields: Operation Identifier: The unique ID and timestamp of this operation; Target Data Product ID and Field List: The data product and specific fields requested by the Agent; Operation Type: The type of operation requested by the Agent (such as "query", "aggregate statistics", "feature extraction", etc.); Operation Parameters: Including specific parameters such as query conditions, aggregation dimensions, and filtering rules; Result Intended Use Declaration: A description of the Agent's intended use of the query result, automatically generated by the Agent's task planning module based on the current inference context (such as "used to calculate regional signal coverage statistics"); Upstream Call Chain Identifier: If this operation is a subsequent operation based on the previous query result, the identifier of the upstream operation is recorded to form an operation chain.

[0051] S42. Based on the constraint description file, perform compliance checks on the intent declaration for data range, operation type, query parameters, and cache expiration time. Calculate the compliance score by weighted averaging the compliance check results.

[0052] In this embodiment, based on the constraint description file, the intent declaration is verified from four dimensions: data range, operation type, query parameters, and cache validity. The verification results are weighted and averaged to obtain a compliance score. Through multi-dimensional fine-grained verification, the system comprehensively checks whether the agent's operation has exceeded the limits. The quantitative score intuitively distinguishes the severity of the violation, providing a standardized quantitative judgment basis for subsequent differentiated graded responses, and realizing precise quantitative control over the agent's autonomous data behavior.

[0053] Specifically, policy interception and real-time comparison are performed. After receiving an intent declaration, the policy interception layer compares it in real-time with the currently active capability envelope. The comparison process follows these rules: Data Scope Check: Verifies that the target data product ID and field list in the intent statement are within the list of accessible data products in the capability envelope. If the agent requests access to an unauthorized data product or field, the operation is rejected.

[0054] Operation type check: Verify whether the operation type in the intent statement is within the operation type whitelist of the capability envelope. For example, if the capability envelope only allows "aggregate statistics" operations, and the Agent requests to execute "query by query", then the operation is rejected.

[0055] Parameter constraint check: Verify whether the operation parameters meet the single query constraints in the capability envelope. For example, check whether the number of returned records exceeds the maximum limit, whether the aggregation granularity is lower than the minimum requirement, and whether the query frequency exceeds the upper limit.

[0056] Cache constraint check: Checks if there are any historical query results in the Agent runtime environment that exceed the cache duration limit. If so, forcibly clears the expired cache before allowing new operations.

[0057] Before executing the tiered response, the policy interception layer calculates the intent compliance score based on the inspection results of each dimension, as shown in the following formula:

[0058] in, For the j-th intent declaration, E is the currently effective capability envelope, and D is the total number of inspection dimensions (data range, operation type, parameter constraints, and cache constraints, a total of 4 dimensions). Let ∑ be the weight coefficient of the d-th dimension. = 1 (assuming equal weights for all dimensions) = 0.25), The result of the inspection for the d-th dimension is categorized by value: 1 (fully compliant), 0 (partially non-compliant, subject to downgrade), and -1 (serious violation). The tiered response rule is: when... Release when = 1.0; release when 0 < When the value is less than 1.0 and no dimension has a value of -1, the process is downgraded; when any dimension's Checked value is -1, the process is blocked and an alarm is triggered; when... Manual intervention is triggered when the value is less than 1.0 but all items exceeding the limit are within the negotiable range.

[0059] S43. When the compliance score determines that the operation is compliant, the data operation is allowed; when the compliance score determines that the operation is slightly out of bounds, the operation level of the data operation is downgraded; when the compliance score determines that the operation is in violation, the data operation is stopped and an alarm is issued.

[0060] In this embodiment, three processing logics are divided based on compliance scores: direct release for compliance, automatic downgrading for minor violations, and blocking and alarming for serious violations. Differentiated graded processing is implemented for different degrees of boundary violations by the intelligent agent. Compliance operations do not require manual intervention to ensure the efficiency of autonomous analysis by the intelligent agent, minor violations are automatically adjusted to avoid task interruption, and high-risk violations are intercepted in real time to stop losses, thus balancing the needs of automated business and the requirements of data compliance management.

[0061] In this embodiment, a tiered response is executed based on the comparison results: Allow: When the intent statement is entirely within the capability envelope, the operation request is automatically allowed without any manual intervention, and the Agent can autonomously complete the data access; Degrade: When the intent statement partially exceeds the capability envelope but is subject to automatic adjustment, the interception layer automatically downgrades the operation before allowing it to proceed. For example, if the Agent requests the return of complete customer records, the interception layer automatically replaces sensitive fields in the returned results with de-identified values, or automatically converts the request's line-by-line query into an aggregated statistical query; Block and Alarm: When the intent statement clearly violates the capability envelope constraints (such as requesting access to unauthorized data products or requesting the export of all raw data), the operation is immediately blocked, the system records the blocking event, and sends an alarm to the task administrator.

[0062] In this embodiment, downgrading the operation level of data operations can convert detailed queries in data operations into aggregated statistical queries; automatically desensitize and mask data-sensitive fields in the output data corresponding to data operations; and compress the number of data records returned by a single query of data operations to within the limited threshold of the digital contract.

[0063] In this embodiment, the operation downgrade is completed by converting detailed queries into aggregated statistical queries, automatically de-identifying and masking sensitive fields, and compressing the number of data records returned by the query. This strictly adheres to the various usage restrictions of digital contracts, enabling compliant adjustments without terminating the smart agent's tasks, while also reducing the outflow of original sensitive data from the source of the query, effectively reducing the risk of data leakage, and balancing normal business operations with data usage compliance.

[0064] S44. When an intent statement exceeds the scope of the constraint description file, manual approval is required to confirm whether to extend the constraint description file, or to reject the data operation.

[0065] In this embodiment, manual approval is initiated for scenarios where the intent declaration exceeds the scope of the constraint description file. The administrator can choose to temporarily extend the constraint file permissions or directly reject the data operation, balancing the reasonable business expansion needs with the bottom line of data security, avoiding the interruption of normal business analysis due to the solidification of constraints, and leaving a trace of the permission change process to ensure that the out-of-scope operation is auditable throughout the process.

[0066] In this embodiment, when an intent to declare an extension beyond the scope of the capability envelope but is a reasonable business requirement (such as when an agent discovers through analysis that it needs to access related fields not listed in the envelope), the system suspends the operation and sends an approval request to the task administrator. The administrator can choose to temporarily extend the scope of the capability envelope or reject the request.

[0067] Please see Figure 5 , Figure 5 This is the fifth flowchart of a method for autonomous access control of intelligent agent data in a trusted data space, as provided in this disclosure. Figure 5 As shown, the control methods include: S51. Calculate the text similarity and structured field matching degree between the original data and the output data.

[0068] In this embodiment, by calculating the text similarity and structured field matching degree between the original data and the output data, the output content is compared and verified from multiple dimensions. This accurately identifies whether there are problems such as leakage of original plaintext or illegal disclosure of fields, improving the comprehensiveness and accuracy of output compliance detection and providing a reliable basis for subsequent correction of illegal content.

[0069] In this embodiment, before the Agent generates the final output (such as analysis reports, response text, visualization charts, etc.), the system performs a compliance scan on the output content. The scanning process is executed based on the output constraints in the capability envelope, and the checks include at least: original record detection, scanning whether the output text contains content fragments that are highly similar to the original records in the query results (achieved through text similarity algorithms and pattern matching); identifiable information detection, scanning whether the output contains information that can identify a specific individual, device, or location (such as base station numbers accurate to single digits, precise latitude and longitude coordinates, etc.); and accuracy compliance check, verifying whether the numerical results in the output meet the accuracy requirements specified in the capability envelope (such as whether the minimum aggregation granularity limit is violated).

[0070] S52. A data leakage risk value is generated by weighted fusion of text similarity and structured field matching degree.

[0071] In this embodiment, text similarity and structured field matching are weighted and fused to quantify the data leakage risk value. The two types of comparison indicators are integrated into a unified risk assessment standard, which realizes an intuitive quantitative classification of the data leakage risk, making the violation judgment more scientific and objective, and facilitating the rapid differentiation of risk levels and the precise implementation of corresponding control measures.

[0072] In this embodiment, the determination formula for original record detection is as follows:

[0073] in, Output the t-th segment of the text to the Agent. To retrieve the kth original record, For text similarity based on edit distance normalization ( = 1 - EditDistance(Ot, Rk) / max(len(Ot), len(Rk))), where EditDistance is the edit distance, usually referring to the Levenshtein distance, which is a quantitative metric for measuring the difference between two strings, and len(Ot) is... The length of , len(Rk) is The length of the array is given by max(len(Ot), len(Rk)), where max(len(Ot), len(Rk)) is the larger of len(Ot) and len(Rk). The structured field matching degree (number of matching fields / total number of fields) is represented by δ1 = 0.6 and δ2 = 0.4, which are weighting coefficients.

[0074] S53. Determine whether the output data violates the constraint description document and data leakage risk value.

[0075] In this embodiment, the rules in the constraint description file and the data leakage risk value obtained by quantification are combined to comprehensively determine whether the output data is in violation of regulations. By integrating the dual basis of institutional constraints and intelligent risk assessment, the violation judgment standard is made more rigorous and comprehensive, effectively improving the accuracy of compliance review of output content and accurately identifying various data leakage risks.

[0076] In this embodiment, when SimOutput(Ot, Rk) ≥ θleak (default θleak = 0.7), it is determined that the output fragment Ot contains the risk of original record leakage, and a correction operation is triggered.

[0077] S54. When the output data is invalid, replace the data in the output data with aggregated statistical descriptions, or perform generalization processing, or perform rounding processing.

[0078] In this embodiment, after determining that the output data is in violation, multiple methods such as aggregated description, content generalization, and numerical rounding are used for rectification. This flexibly adapts to different types of in violation data to complete compliance correction. While retaining the core business information, sensitive original content is removed without having to discard the output results. This approach safeguards the bottom line of data security while ensuring the normal use of the intelligent agent's business analysis results.

[0079] In this embodiment, for any violations discovered during scanning, the system automatically performs corrective actions: replacing the original record fragments with aggregated descriptions (e.g., replacing "the signal strength of base station A is -75dBm" with "the average signal strength of base stations in this area is -72dBm"); generalizing identifiable information (e.g., replacing precise latitude and longitude with administrative division descriptions); and rounding values ​​exceeding precision limits. Each step of the corrective action is recorded in the audit log.

[0080] Please see Figure 6 , Figure 6 This disclosure provides an intelligent agent autonomous data access control system for a trusted data space. The control system includes: a constraint compilation module 11, an operation response module 12, and a data output module 13.

[0081] In this embodiment, the constraint compilation module 11 is used to obtain data access constraints in the digital contract of the trusted data space, compile the data access constraints, and generate a constraint description file by signing it and sending it to the intelligent agent's runtime environment.

[0082] In this embodiment, the operation response module 12 is used to generate an intent statement for the corresponding data operation when the agent performs data operation on the trusted data space, perform multi-dimensional compliance verification on the intent statement based on the constraint description file to generate a compliance score, and perform graded response to the data operation according to the compliance score.

[0083] In this embodiment, the data output module 13 is used to perform a compliance scan on the output data based on the constraint description file and the original data before the intelligent agent completes the data operation and outputs the data; when the output data is non-compliant, it performs a correction operation on the non-compliant content in the output data; when the output data is compliant, it outputs the output data.

[0084] In this embodiment, the control system further includes: a trajectory generation module, used to obtain intent declaration, compliance score, response type of graded response, data operation, output data, operation time of data operation, and resource consumption; link the intent declaration, compliance score, response type of graded response, data operation, output data, operation time of data operation, and resource consumption in chronological order and calculate the hash value of the event node to obtain a behavior trajectory chain; associate the digital contract with the behavior trajectory chain, and generate a compliance proof report for the corresponding digital contract based on the operation status of the data operation, the actual execution status of each data access constraint in the constraint description file, and the scan results of the compliance scan.

[0085] In this embodiment, the constraint compilation module 11 is specifically used to extract the contract target, operation blacklist / whitelist, time constraint, frequency constraint, environmental constraint, and obligation clause constraint from the digital contract in the trusted data space; based on the contract target, operation blacklist / whitelist, time constraint, frequency constraint, environmental constraint, and obligation clause constraint, it compiles and generates a capability envelope containing a unique envelope identifier, authorized data products and fields, operation whitelist, single query limit, output desensitization rules, cache expiration, and validity period; it signs the capability envelope based on the data provider's key to generate a constraint description file, and distributes the constraint description file to the intelligent agent's runtime environment.

[0086] In this embodiment, the operation response module 12 is specifically used to generate an intent declaration containing an operation identifier, accessed data source, operation type, query parameters, business purpose, and upstream call chain when the intelligent agent performs data operations on the trusted data space; based on the constraint description file, the intent declaration is subjected to compliance checks on data range, operation type, query parameters, and cache expiration timeliness, and a weighted average compliance score is calculated on the compliance check results; when the compliance score determines that the operation is compliant, the data operation is allowed; when the compliance score determines that the operation is slightly excessive, the operation level of the data operation is downgraded; when the compliance score determines that the operation is illegal, the data operation is stopped and an alarm is issued; when the intent declaration exceeds the scope of the constraint description file, manual approval is performed to confirm whether to expand the constraint description file or reject the data operation.

[0087] In this embodiment, the operation response module 12 is specifically used to convert the detailed query in the data operation into an aggregated statistical query; automatically desensitize and mask the data-sensitive fields of the output data corresponding to the data operation; and compress the number of data entries returned by a single query of the data operation to within the limited threshold of the digital contract.

[0088] In this embodiment, the data output module 13 is specifically used to calculate the text similarity and structured field matching degree of the original data and the output data; to generate a data leakage risk value by weighted fusion of the text similarity and structured field matching degree; to determine whether the output data violates the rules based on the constraint description file and the data leakage risk value; and when the output data violates the rules, to replace the data in the output data with aggregated statistical description, or to perform generalization processing, or to perform rounding processing.

[0089] In this embodiment, the constraint compilation module 11 is also used to add a globally unique ID to the constraint description file and associate the globally unique ID with the digital contract and data operation.

[0090] like Figure 7As shown, this embodiment of the present disclosure provides an electronic device, including a processor 1110, a communication interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. Memory 1130 is used to store computer programs; When the processor 1110 executes the program stored in the memory 1130, it implements any of the above methods.

[0091] The electronic device provided in this embodiment includes a processor 1110 that executes a program stored in a memory 1130 to obtain data access constraints in a digital contract within a trusted data space. The processor compiles and signs the data access constraints to generate a constraint description file, which is then sent to the intelligent agent's runtime environment. When the intelligent agent performs data operations on the trusted data space, it generates an intent statement for the corresponding data operation. Based on the constraint description file, it performs multi-dimensional compliance verification on the intent statement to generate a compliance score, and provides a graded response to the data operation based on the compliance score. Before the intelligent agent completes the data operation and outputs the data, it performs a compliance scan on the output data based on the constraint description file and the original data. When the output data is found to be non-compliant, the non-compliant content in the output data is corrected. When the output data is compliant, the output data is output.

[0092] The communication bus 1140 mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, and a component bus, etc. For ease of illustration, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus.

[0093] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0094] The memory 1130 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory 1130 may also be at least one storage device located remotely from the aforementioned processor 1110.

[0095] The processor 1110 mentioned above can be a general-purpose processor 1110, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0096] This disclosure provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors 1110 to implement the methods of any of the above embodiments.

[0097] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this disclosure is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0098] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for autonomous access control of intelligent agent data in a trusted data space, characterized in that, The control method includes: Obtain data access constraints from the digital contracts in the trusted data space, compile and sign the data access constraints to generate a constraint description file, and send it to the intelligent agent's runtime environment; When the intelligent agent performs data operations on the trusted data space, it generates an intent statement corresponding to the data operation, performs multi-dimensional compliance verification on the intent statement based on the constraint description file to generate a compliance score, and responds to the data operation in a graded manner according to the compliance score. Before the agent completes data operations and outputs data, a compliance scan is performed on the output data based on the constraint description file and the original data; if the output data violates the rules, the violation content in the output data is corrected; if the output data is compliant, the output data is output.

2. The data discretionary access control method according to claim 1, characterized in that, The control method further includes: Obtain the intent statement, the compliance score, the response type of the graded response, the data operation, the output data, the operation time of the data operation, and the resource consumption; The intent statement, compliance score, response type of graded response, data operation, output data, operation time and resource consumption of the data operation are linked in chronological order, and the hash value is calculated according to the event node to obtain the behavior trajectory chain; The digital contract is associated with the behavior trajectory chain, and a compliance proof report corresponding to the digital contract is generated based on the operation status of the data operation, the actual execution status of each data access constraint in the constraint description file, and the scanning results of the compliance scan.

3. The data discretionary access control method according to claim 1, characterized in that, The process of obtaining data access constraints from digital contracts in the trusted data space, compiling and signing the data access constraints to generate a constraint description file, and then sending it to the agent's runtime environment includes: Extract the contract target, operation blacklist / whitelist, time constraint, frequency constraint, environmental constraint, and obligation clause constraint from the digital contracts in the trusted data space; Based on the contract subject matter, operation blacklist / whitelist, time constraints, frequency constraints, environmental constraints, and obligation clause constraints, a capability envelope containing a unique envelope identifier, authorized data products and fields, operation whitelist, single query limit, output de-identification rules, cache expiration time, and validity period is compiled and generated. The capability envelope is signed using the data provider's key to generate a constraint description file, which is then sent to the agent's runtime environment.

4. The data discretionary access control method according to claim 1, characterized in that, When the agent performs data operations on the trusted data space, it generates an intent statement corresponding to the data operation, performs multi-dimensional compliance verification on the intent statement based on the constraint description file to generate a compliance score, and provides a graded response to the data operation based on the compliance score, including: When the intelligent agent performs data operations on the trusted data space, it generates an intent declaration that includes an operation identifier, accessed data source, operation type, query parameters, business purpose, and upstream call chain. Based on the constraint description file, the intent declaration is subjected to compliance checks for data range, operation type, query parameters, and cache expiration time. The compliance score is calculated by weighted average of the compliance check results. When the compliance score determines that the operation is compliant, the data operation is allowed; when the compliance score determines that the operation is slightly out of bounds, the operation level of the data operation is downgraded; when the compliance score determines that the operation is in violation, the data operation is stopped and an alarm is issued. When the intent statement exceeds the scope of the constraint description file, manual review is required to determine whether to extend the constraint description file or reject the data operation.

5. The data discretionary access control method according to claim 4, characterized in that, The downgrading of the operation level for the data operation includes: Convert the detailed query in the data operation into an aggregated statistical query; The data-sensitive fields of the output data corresponding to the data operations are automatically de-identified and masked. The number of data entries returned in a single query of the data operation is compressed to within the limited threshold of the digital contract.

6. The data discretionary access control method according to claim 1, characterized in that, The output data is then subjected to a compliance scan based on the constraint description file and the original data. When the output data is invalid, a correction operation is performed on the invalid content in the output data, including: Calculate the text similarity and structured field matching degree between the original data and the output data; A data leakage risk value is generated by weighted fusion of the text similarity and the structured field matching degree; Determine whether the output data violates the constraint description file and the data leakage risk value; When the output data is invalid, the data in the output data is replaced with an aggregated statistical description, or generalized, or rounded.

7. The data discretionary access control method according to any one of claims 1 to 6, characterized in that, The control method further includes: Add a globally unique ID to the constraint description file, and associate the globally unique ID with the digital contract and the data operation.

8. A data autonomous access control system for intelligent agents oriented towards a trusted data space, characterized in that, The control system includes: The constraint compilation module is used to obtain data access constraints in the digital contracts of the trusted data space, compile and sign the data access constraints to generate a constraint description file and send it to the intelligent agent's runtime environment. The operation response module is used to generate an intent statement corresponding to the data operation when the agent performs a data operation on the trusted data space, perform multi-dimensional compliance verification on the intent statement based on the constraint description file to generate a compliance score, and perform a graded response to the data operation according to the compliance score. The data output module is used to perform a compliance scan on the output data based on the constraint description file and the original data before the intelligent agent completes the data operation and outputs the data; when the output data is non-compliant, it corrects the non-compliant content in the output data; when the output data is compliant, it outputs the output data.

9. An electronic device, characterized in that, include: processor; as well as A memory storing computer-executable instructions, which, when executed, cause the processor to perform the control method according to any one of claims 1-7.

10. A computer storage medium, characterized in that, in, The computer storage medium stores one or more programs, which, when executed by a processor, implement the control method according to any one of claims 1-7.

Images