Intention evidence, risk classification, and state freshness agent access control method

Abstract

The application discloses an agent access control method of intention evidence, risk grading and state freshness, acquires and verifies registered, task and runtime intention evidence, maps effective evidence to a versioned business purpose directory, and determines missing evidence. According to action, resource, parameter, data level, credential range and previous action, target drift and action risk level are calculated, and a freshness vector of policy, identity binding, credential revocation, agent revocation and certificate state is generated according to the same, and a real-time query, local calculation or cache reuse path is selected. A signature policy package is verified and a cache key is constructed, and a local decision is compared with an event water level, a heartbeat lease and a freshness vector before use; a high-risk action is forced to acquire a current policy and a real-time revocation state. The application outputs release, supplementary evidence, limited rights, suspension, isolation or rejection results based on intention, drift, risk and security state; when a state anomaly or a failure event is discontinuous, a cache is frozen and a failure is closed.

Description

Technical Field

[0001] This invention relates to the fields of agent security, access control, and distributed policy enforcement, and more specifically, to an agent access control method based on intent evidence, risk classification, and state freshness. Background Technology

[0002] Intelligent agents can invoke tools, application programming interfaces (APIs), model context protocols (MCPs), services, and data resources. Traditional access control primarily determines whether a subject has static permissions, making it difficult to determine whether the current invocation still aligns with the approved business purpose.

[0003] Existing intent recognition is vulnerable to prompt injection, task forgery, and target drift; distributed policy nodes may also continue to use expired policies, credential states, or revoked states. Even if the intent is correctly determined, expired security states may still lead to erroneous permission granting. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an agent access control method based on intent evidence, risk classification and state freshness, which can put credible intent evidence, action risk, policy version and security state freshness into the same runtime decision, so that the decision basis is credible and sufficiently new, and blocks the call when the necessary state cannot be confirmed.

[0005] The technical solution adopted by this invention to solve its technical problem is: constructing an agent access control method based on intent evidence, risk classification, and state freshness, comprising the following steps: S1. Obtain the registration intent evidence of the target intelligent agent, the task intent evidence of this task, and the runtime intent evidence when the tool call occurs, respectively; S2. Verify the source identity, digital signature, authorization chain, object version, and time validity of registration intent evidence, task intent evidence, and runtime intent evidence; map valid evidence to versioned business purpose directory; and form a set of missing evidence. S3. Calculate the target drift based on the current action, resources, parameters, data level, credential range, and previous action sequence, and determine the action risk level by combining resource sensitivity level, permission impact, and environment. S4. Based on the target drift amount and action risk level, generate freshness vector components for policy, identity binding, credential revocation, agent revocation and certificate status according to the versioned mapping rules. Select the central real-time query, local policy calculation or decision cache reuse path based on whether all necessary freshness vector components are satisfied. S5. Obtain the policy package signed by the policy publisher using the private key corresponding to the PKI certificate, verify the policy package signature, certificate status, release version, field pattern, validity period and algorithm policy, and construct the binding intent evidence digest, destination directory version and cache keys for various security status versions. S6. Before using local decision-making or cached results, compare the local processed event water level, authority high water level, signature heartbeat lease, and freshness vector; for high-risk actions, force the acquisition of the current policy and real-time revocation status, and do not use historical decision cache; S7. The multi-layered intent state, target drift amount, action risk level and safety state that meets the freshness requirement are formed into policy inputs, and the results of release, supplementary certification, restriction of authority, suspension, isolation or rejection are generated and executed by the runtime control node before the target tool receives the request. S8. Issue failure events with sequential numbers when policies, certificates, identity bindings, credentials, or agent states change; freeze relevant caches and close them upon failure when invalid evidence, missing sequence numbers, lagging water levels, lease timeouts, or unknown necessary states occur.

[0006] According to the above scheme, in step S1, the registration intent evidence, task intent evidence, and runtime intent evidence each have a source identifier, object version, generation time, and validity period. The registration intent comes from the agent registration center, the task intent comes from the work order or task orchestration system, and the runtime intent is formed by the control node near the tool entry point. The registration intent evidence, task intent evidence, and runtime intent evidence use independent versions.

[0007] According to the above scheme, in step S2, the evidence verification process handles the deterministic fields and password verification, and the semantic model only outputs the normalized risk attributes, without covering signature, authorization chain and time limit failure. The business purpose directory records the purpose number, allowed actions, resource type, data range, validity period, and directory version. At runtime, the registration intent evidence, task intent evidence, and runtime intent evidence are mapped to the valid purpose number and then participate in the target drift calculation.

[0008] According to the above scheme, in step S3, the target drift amount is calculated based on at least two of the semantic distance, resource boundary crossing degree, credential range deviation and preceding action change rate, according to versioned combination rules; the target drift and risk level adopt a versioned field mode, and unknown fields generate supplementary certification, suspension or rejection results according to risk strategy.

[0009] According to the above scheme, the method for calculating the target drift amount according to the versioned combination rule is as follows: normalize the semantic distance between the current call relative registration and the task purpose, the degree of resource boundary violation, the credential range deviation and the change rate of the preceding action to the interval of 0 to 1 respectively, and then combine them into drift components according to the weight of the version with mode. The semantic distance is obtained by the vector similarity or rule matching degree between the call description and the versioned business purpose directory entry, and the preceding action change rate is obtained by the difference rate between the preceding action sequence and the historical benchmark sequence; any normalized component exceeding the corresponding threshold or the combined drift value exceeding the drift threshold is judged as a deviation.

[0010] According to the above scheme, in step S4, each component of the freshness vector includes a safety status category, maximum allowable age, minimum event level, query mode, and failure handling method; the versioned mapping rule indexes the corresponding components according to the target drift range and action risk level, the maximum allowable age corresponding to a higher drift or higher risk level is not greater than the lower level, the minimum event level is not lower than the lower level, and the query mode cannot be relaxed from real-time query to cache reuse.

[0011] According to the above scheme, in step S5, the strategy package includes tenant, strategy version, field pattern, algorithm strategy, validity period and publisher information, and the cache key is simultaneously bound to the evidence and status version that affect the authorization result.

[0012] According to the above scheme, in step S6, the validity period of the cached result shall not exceed the minimum value among the validity period of the policy package, the validity period of the intent evidence, the validity period of the attribute snapshot, the validity period of the revocation status, and the signature heartbeat lease; high-risk calls bypass the historical decision cache, and low-risk calls reuse the result only when the water level, lease, and all necessary states meet the requirements.

[0013] According to the above scheme, in step S8, the intent evidence or policy package is signed by the corresponding subject using the private key corresponding to its PKI certificate. The digital signature is implemented based on classical cryptographic algorithms such as SM2, RSA, and ECC, as well as post-quantum algorithms. The signature certificate status, key version, and algorithm policy version are incorporated into the policy input or cache invalidation conditions.

[0014] This invention also provides an intelligent agent access control system based on multi-layered intent evidence, risk classification, and state freshness, comprising, The intent evidence collector module is used to obtain and record independent versions of registration, task and runtime intent evidence. The output data includes object identifier, generation time, mode version and source identifier, which are used for subsequent modules to verify and associate. The Evidence Verification and Drift Analyzer module is used to verify evidence and calculate purpose matching and target drift. The output data includes object identifier, generation time, pattern version and source identifier, which are used for subsequent modules to verify and correlate. The Risk and Freshness Calculator module is used to determine the risk of an action and the freshness requirements of various safety states. The output data includes the object identifier, generation time, mode version, and source identifier, which are used for subsequent modules to verify and associate. The PKI policy package service module is used to sign and publish policy packages and provide certificates and algorithm policy status. The output data includes object identifier, generation time, mode version and source identifier, which are used for subsequent modules to verify and associate. The distributed policy executor module is used to select decision paths, verify caches, and execute control results. The output data includes object identifiers, generation times, mode versions, and source identifiers for subsequent modules to verify and associate. The Failure Synchronization and Action Chain Service module is used to propagate failure events, maintain the authority level, and save the action chain. The output data includes object identifier, generation time, mode version, and source identifier for subsequent modules to verify and associate. The PKI policy package service module provides authentication, digital signature, certificate status, and trusted time capabilities for intent evidence, policy packages, or state packages. Depending on the deployment environment and cryptographic compliance requirements, the relevant signature and verification capabilities are implemented based on classical cryptographic algorithms such as SM2, RSA, and ECC, as well as post-quantum algorithms, and the algorithm policy version is recorded to support smooth migration.

[0015] The agent access control method based on intent evidence, risk classification, and state freshness of the present invention has the following beneficial effects: This invention combines "whether the current call meets the task objective" and "whether the policy and status used for judgment are sufficiently new" into a single runtime security decision. It can identify calls with valid static permissions but whose objectives have deviated from the target, and avoids allowing calls to proceed using expired policies or revoked statuses. High-risk actions are handled using real-time status, while low-risk actions use local policies or caching under clear freshness conditions, balancing security and execution latency. The decision record includes evidence, risk, version, water level, and failure reason, facilitating auditing and policy regression. Attached Figure Description

[0016] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a structural diagram of the intelligent agent access control system based on multi-layered intent evidence, risk classification, and state freshness of the present invention. Figure 2 This is a flowchart of an agent access control method based on the present invention's intent evidence, risk classification, and state freshness. Detailed Implementation

[0017] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] like Figure 1-2 As shown, the agent access control method of the present invention based on intent evidence, risk classification, and state freshness specifically includes the following steps: S1. Obtain the registration intent evidence of the target intelligent agent, the task intent evidence of this task, and the runtime intent evidence when the tool is invoked; the registration intent evidence, task intent evidence, and runtime intent evidence each have a source identifier, object version, generation time, and validity period.

[0019] The registration intent evidence is provided by the agent registry center, the task intent evidence is provided by the task orchestration or work order system, and the runtime intent evidence is formed by the tool entry control node. These three types of evidence use independent versions. The necessity of the registration intent evidence, task intent evidence, and runtime intent evidence is determined based on the action risk level. If the signature, authorization chain, or timeliness verification of any necessary evidence fails, it cannot be offset by other layers of evidence or models. Policies, identity binding, credential revocation, agent revocation, and certificate status are obtained from their respective authoritative services. If any necessary evidence, status version, or authority level cannot be verified, the runtime control node generates a supplementary verification, suspension, or rejection result according to the risk level, without using historical decisions whose freshness cannot be confirmed.

[0020] In step S2, the source identity, digital signature, authorization chain, object version, and time validity of the registration intent evidence, task intent evidence, and runtime intent evidence are verified. Valid evidence is mapped to the versioned business purpose directory, and a set of missing evidence is formed.

[0021] Evidence verification first processes deterministic fields and password verification. The semantic model only outputs normalized risk attributes and does not cover signature, authorization chain, and expiration failure. The business purpose directory records at least the purpose number, allowed actions, resource type, data range, validity period, and directory version. Runtime intents only participate in target drift calculations after being mapped to a valid purpose number.

[0022] The necessity of registration intent evidence, task intent evidence, and runtime intent evidence is determined based on the action risk level. If the signature, authorization chain, or timeliness verification of any necessary evidence fails, it shall not be offset by other layers of evidence or models.

[0023] In step S3, the target drift is calculated based on the current action, resources, parameters, data level, credential range, and previous action sequence, and the action risk level is determined by combining the resource sensitivity level, permission impact, and environment.

[0024] The target drift is calculated based on at least two of the following: semantic distance, resource out-of-bounds degree, credential range deviation, and preceding action change rate, according to a versioned combination rule. Target drift and risk level adopt a versioned field pattern, and unknown fields generate supplementary certification, suspension, or rejection results according to the risk strategy.

[0025] The target drift amount is calculated using versioned combination rules as follows: the semantic distance between the current call and the registration and task objective, the degree of resource boundary violation, the credential range deviation, and the change rate of preceding actions are normalized to the interval between 0 and 1, and then combined into drift components according to the weights of the pattern version. The semantic distance is obtained from the vector similarity or rule matching degree between the call description and the versioned business objective directory entry, and the change rate of preceding actions is obtained from the difference rate between the preceding action sequence and the historical baseline sequence. Any deviation is considered when any normalized component exceeds the corresponding threshold or when the combined drift value exceeds the drift threshold. The normalization method, weights, thresholds, and combination rules are all fixed according to the field pattern version, and can be recalculated accordingly during evidence verification.

[0026] In step S4, based on the target drift amount and action risk level, freshness vector components are generated for policy, identity binding, credential revocation, agent revocation and certificate status according to the versioned mapping rules; depending on whether all necessary components are satisfied, the central real-time query, local policy calculation or decision cache reuse path is selected.

[0027] The freshness vector component must include at least the safety status category, maximum allowed age, minimum event level, query mode, and failure handling method, and the corresponding requirements must not be relaxed when the target drift amount or action risk level increases; the freshness vector is calculated separately for each safety status category, so that low-risk calls cannot use a newer type of status to cover up an expired type of status.

[0028] Versioned mapping rules employ a decision table jointly indexed by the target drift range and action risk level. For each safety state, it outputs the maximum permissible age, minimum event threshold, query mode, and failure handling method. Versioned mapping rules index corresponding components according to the target drift range and action risk level. The maximum permissible age for a higher drift range or risk level cannot exceed that of the lower level, and the minimum event threshold cannot be lower than that of the lower level. Furthermore, the query mode cannot be relaxed from real-time query to cache reuse. At runtime, all necessary components are checked item by item. Real-time queries for any component cannot be replaced by local caching, and historical decisions cannot be reused if any component fails to meet the requirements.

[0029] In step S5, the policy package signed by the policy publisher using the private key corresponding to the PKI certificate is obtained. The policy package signature, certificate status, release version, field pattern, validity period and algorithm policy are verified. The cache keys for binding intent evidence digest, destination directory version and various security status versions are constructed.

[0030] The policy package includes tenant, policy version, field schema, algorithm policy, validity period, and publisher information. The cache key is also bound to the evidence and status version that affect the authorization result. The policy package release version monotonically increments, and when restoring historical policy content, a newer and higher release version is used, and an expiration event is generated.

[0031] S6. Before using local decision-making or cached results, compare the local processed event water level, authority high water level, signature heartbeat lease, and freshness vector; for high-risk actions, force the acquisition of the current policy and real-time revocation status, and do not use historical decision cache.

[0032] High-risk calls bypass the historical decision cache; low-risk calls reuse results only when the water level, lease, and all necessary states meet the requirements. The validity period of the cached results does not exceed the minimum of the policy package validity period, intent evidence validity period, attribute snapshot validity period, revocation status validity period, and signature heartbeat lease validity period.

[0033] S7, takes the multi-layered intent state, target drift amount, action risk level and safety state that meets the freshness requirement as policy input, generates release, supplementary certification, restriction of authority, suspension, isolation or rejection results, and executes them by the runtime control node before the target tool receives the request.

[0034] The runtime control node makes decisions before sending them to the target tool and saves a summary of the basis for the decisions, but does not save the original intermediate output of the model that is not related to control. When an event sequence number gap, water level lag, or signature heartbeat timeout is detected, the node stops hitting the affected cache, completes state resynchronization, and restores the system after verifying the authority water level.

[0035] In step S8, failure events with consecutive serial numbers are published when the policy, certificate, identity binding, credentials, or agent state changes; when invalid evidence, missing serial numbers, lagging water level, lease timeout, or unknown necessary state occurs, the relevant cache is frozen and shut down upon failure.

[0036] Failure events propagate sequentially. When the authoritative level cannot be confirmed, the cache remains frozen to prevent network failures from being interpreted as unchanged security status. Intent evidence or policy packets are signed by the corresponding entity using the private key corresponding to its PKI certificate. Digital signatures are implemented based on classic cryptographic algorithms such as SM2, RSA, and ECC, as well as post-quantum algorithms. The signature certificate status, key version, and algorithm policy version are incorporated into the policy input or cache failure conditions. Example

[0037] An operations and maintenance agent is registered to read the security configuration of a specified environment. This task requires checking a set of protected resources. The agent then requests to modify the access policy. After verifying the registration, task, and runtime intent, the system finds that the action is outside the scope of the task and judges it as high-risk. At the same time, the local node's credential revocation level is lagging behind the authority level, so the system does not use the cached result, directly rejects the call, and records the evidence status and level difference.

[0038] This invention also provides an agent access control system for intent evidence, risk classification, and state freshness, including an intent evidence collector module, an evidence verification and drift analyzer module, a risk and freshness calculator module, a PKI policy package service module, a distributed policy executor module, and a failure synchronization and action chain service module. The modules exchange data through interfaces with authentication and integrity protection. The system can be deployed on the same management platform, or the runtime modules can be deployed close to the agent or target resource, while the control, evidence, and management modules can be deployed on a management network.

[0039] The system comprises the following modules: **Intent Evidence Collector:** This module acquires registration, task, and runtime intent evidence and records independent versions. The output data includes object identifier, generation time, mode version, and source identifier for subsequent modules to verify and correlate. **Evidence Verification and Drift Analyzer:** This module verifies evidence and calculates purpose matching and target drift. The output data includes object identifier, generation time, mode version, and source identifier for subsequent modules to verify and correlate. **Risk and Freshness Calculator:** This module determines action risks and freshness requirements for various security states. The output data includes object identifier, generation time, mode version, and source identifier for subsequent modules to verify and correlate. **PKI Policy Package Service:** This module signs and publishes policy packages and provides certificates and algorithm policy states. The output data includes object identifier, generation time, mode version, and source identifier for subsequent modules to verify and correlate. **Distributed Policy Executor:** This module selects decision paths, verifies caches, and executes control results. The output data includes object identifier, generation time, mode version, and source identifier for subsequent modules to verify and correlate. The Failure Synchronization and Action Chain Service module is used to propagate failure events, maintain authority levels, and preserve the action chain. The output data includes object identifiers, generation times, mode versions, and source identifiers for subsequent module verification and correlation. The PKI Policy Package Service module provides authentication, digital signatures, certificate status, and trusted time capabilities for intent evidence, policy packages, or state packages. Depending on the deployment environment and cryptographic compliance requirements, the relevant signature and verification capabilities are implemented based on classic cryptographic algorithms such as SM2, RSA, and ECC, as well as post-quantum algorithms, and the algorithm policy version is recorded to support smooth migration.

[0040] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. An agent access control method based on intent evidence, risk classification, and state freshness, characterized in that, Includes the following steps: S1. Obtain the registration intent evidence of the target intelligent agent, the task intent evidence of this task, and the runtime intent evidence when the tool call occurs, respectively; S2. Verify the source identity, digital signature, authorization chain, object version, and time validity of registration intent evidence, task intent evidence, and runtime intent evidence; map valid evidence to versioned business purpose directory; and form a set of missing evidence. S3. Calculate the target drift based on the current action, resources, parameters, data level, credential range, and previous action sequence, and determine the action risk level by combining resource sensitivity level, permission impact, and environment. S4. Based on the target drift amount and action risk level, generate freshness vector components for policy, identity binding, credential revocation, agent revocation and certificate status according to the versioned mapping rules. Select the central real-time query, local policy calculation or decision cache reuse path based on whether all necessary freshness vector components are satisfied. S5. Obtain the policy package signed by the policy publisher using the private key corresponding to the PKI certificate, verify the policy package signature, certificate status, release version, field pattern, validity period and algorithm policy, and construct the binding intent evidence digest, destination directory version and cache keys for various security status versions. S6. Before using local decision-making or cached results, compare the local processed event water level, authority high water level, signature heartbeat lease, and freshness vector; for high-risk actions, force the acquisition of the current policy and real-time revocation status, and do not use historical decision cache; S7. The multi-layered intent state, target drift amount, action risk level and safety state that meets the freshness requirement are formed into policy inputs, and the results of release, supplementary certification, restriction of authority, suspension, isolation or rejection are generated and executed by the runtime control node before the target tool receives the request. S8. Issue failure events with sequential numbers when policies, certificates, identity bindings, credentials, or agent states change; freeze relevant caches and close them upon failure when invalid evidence, missing sequence numbers, lagging water levels, lease timeouts, or unknown necessary states occur.

2. The agent access control method based on intent evidence, risk classification, and state freshness according to claim 1, characterized in that, In step S1, the registration intent evidence, task intent evidence, and runtime intent evidence each have a source identifier, object version, generation time, and validity period. The registration intent comes from the agent registration center, the task intent comes from the work order or task orchestration system, and the runtime intent is formed by the control node near the tool entry point. The registration intent evidence, task intent evidence, and runtime intent evidence use independent versions.

3. The agent access control method based on intent evidence, risk classification, and state freshness according to claim 1, characterized in that, In step S2, the evidence verification process handles deterministic fields and password verification. The semantic model only outputs normalized risk attributes and does not cover signature, authorization chain and timeout failure. The business purpose directory records the purpose number, allowed actions, resource type, data range, validity period, and directory version. At runtime, the registration intent evidence, task intent evidence, and runtime intent evidence are mapped to the valid purpose number and then participate in the target drift calculation.

4. The agent access control method based on intent evidence, risk classification, and state freshness according to claim 1, characterized in that, In step S3, the target drift is calculated according to at least two of the following: semantic distance, resource boundary crossing degree, credential range deviation, and preceding action change rate, based on versioned combination rules. The target drift and risk level adopt a versioned field pattern, and unknown fields generate supplementary certification, suspension, or rejection results according to the risk strategy.

5. The agent access control method based on intent evidence, risk classification, and state freshness according to claim 4, characterized in that, The method for calculating the target drift amount according to the versioned combination rule is as follows: normalize the semantic distance between the current call relative to the registration and the task objective, the degree of resource boundary violation, the credential range deviation and the change rate of the preceding action to the interval of 0 to 1 respectively, and then combine them into drift components according to the weight of the version with mode. The semantic distance is obtained by the vector similarity or rule matching degree between the call description and the versioned business purpose directory entry, and the preceding action change rate is obtained by the difference rate between the preceding action sequence and the historical benchmark sequence; any normalized component exceeding the corresponding threshold or the combined drift value exceeding the drift threshold is judged as a deviation.

6. The agent access control method based on intent evidence, risk classification, and state freshness according to claim 1, characterized in that, In step S4, each component of the freshness vector includes a safety status category, maximum allowed age, minimum event level, query mode, and failure handling method. The versioned mapping rule indexes the corresponding components according to the target drift range and action risk level. The maximum allowed age corresponding to a higher drift or higher risk level is not greater than the lower level, the minimum event level is not lower than the lower level, and the query mode cannot be relaxed from real-time query to cache reuse.

7. The agent access control method based on intent evidence, risk classification, and state freshness according to claim 1, characterized in that, In step S5, the policy package includes tenant, policy version, field pattern, algorithm policy, validity period and publisher information, and the cache key is simultaneously bound to the evidence and status version that affect the authorization result.

8. The agent access control method based on intent evidence, risk classification, and state freshness according to claim 1, characterized in that, In step S6, the validity period of the cached result shall not exceed the minimum value among the validity period of the policy package, the validity period of the intent evidence, the validity period of the attribute snapshot, the validity period of the revocation status, and the signature heartbeat lease. High-risk calls bypass the historical decision cache, while low-risk calls reuse results only when the water level, lease, and all necessary states meet the requirements.

9. The agent access control method based on intent evidence, risk classification, and state freshness according to claim 1, characterized in that, In step S8, the intent evidence or policy package is signed by the corresponding subject using the private key corresponding to its PKI certificate. The digital signature is implemented based on classical cryptographic algorithms such as SM2, RSA, and ECC, as well as post-quantum algorithms. The signature certificate status, key version, and algorithm policy version are incorporated into the policy input or cache invalidation conditions.

10. An agent access control system based on intent evidence, risk classification, and state freshness, characterized in that, include, The intent evidence collector module is used to obtain and record independent versions of registration, task and runtime intent evidence. The output data includes object identifier, generation time, mode version and source identifier, which are used for subsequent modules to verify and associate. The Evidence Verification and Drift Analyzer module is used to verify evidence and calculate purpose matching and target drift. The output data includes object identifier, generation time, pattern version and source identifier, which are used for subsequent modules to verify and correlate. The Risk and Freshness Calculator module is used to determine the risk of an action and the freshness requirements of various safety states. The output data includes the object identifier, generation time, mode version, and source identifier, which are used for subsequent modules to verify and associate. The PKI policy package service module is used to sign and publish policy packages and provide certificates and algorithm policy status. The output data includes object identifier, generation time, mode version and source identifier, which are used for subsequent modules to verify and associate. The distributed policy executor module is used to select decision paths, verify caches, and execute control results. The output data includes object identifiers, generation times, mode versions, and source identifiers for subsequent modules to verify and associate. The Failure Synchronization and Action Chain Service module is used to propagate failure events, maintain the authority level, and save the action chain. The output data includes object identifier, generation time, mode version, and source identifier for subsequent modules to verify and associate. The PKI policy package service module provides authentication, digital signature, certificate status, and trusted time capabilities for intent evidence, policy packages, or state packages. Depending on the deployment environment and cryptographic compliance requirements, the relevant signature and verification capabilities are implemented based on classical cryptographic algorithms such as SM2, RSA, and ECC, as well as post-quantum algorithms, and the algorithm policy version is recorded to support smooth migration.

Images