A government affair scene AI sensitive data access permission management method

Abstract

The application discloses a kind of government affair scene AI sensitive data access permission control method, belong to cross-domain data security permission control technical field.This method is executed between the data access proxy node of mutually independent government affair system and data hosting node, and each other maintains local data permission model and data hierarchical rule in two ends;Two ends complete two-way identity and meta-template negotiation by handshake request message, handshake response message;Four kinds of dimension fields, such as data hosting node is executed boundary clipping according to local security strategy to data classification, operation type, authorization time limit, desensitization level;Two ends save the data permission set after clipping form mutual recognition data authorization snapshot and confirm into effect by hash check;When the change of authorization state, through state change message two-way synchronous update snapshot, simultaneously to interactive whole process structured retention record, it can also be combined block chain and realize that permission snapshot is stored on chain.This application solves the problem that permission rule is not unified when AI accesses sensitive data between heterogeneous government affair system, permission boundary is uncontrollable, and the permission state of two ends is inconsistent, is applicable to the security control scene of multi-department government affair data sharing, AI agent cross-system call government affair sensitive data.

Description

Technical Field

[0001] This invention relates to the fields of government data security, cross-domain permission negotiation, and artificial intelligence data access control, and in particular to a method for controlling access permissions to AI sensitive data in government scenarios. Background Technology

[0002] As e-government continues to deepen, there are increasingly more business scenarios involving cross-departmental data collaboration and AI-powered intelligent agents automatically retrieving cross-departmental government data. Different government departments have built their own independent business systems, each with its own defined data classification standards, executable operation types, access validity periods, and anonymization control requirements, and there is no unified permission interaction protocol.

[0003] Most existing technologies only complete permission approval within a single system. When an external AI agent requests access to sensitive government data across systems, there are significant shortcomings: First, the lack of a standardized handshake negotiation process between the two ends makes it impossible to align the permission definition meta templates in advance, resulting in inconsistent permission parameter formats and high integration costs. Second, access permissions are submitted unilaterally by the applicant, and the receiving end lacks a mechanism to automatically classify data, operation type, access duration, and desensitization strength according to local security policies, which can easily lead to unauthorized access risks. Third, the permission configurations on both ends are stored independently, without forming a permission snapshot that is jointly confirmed by both parties. When permissions change, there is a lack of standardized synchronization messages, which can easily lead to inconsistencies in the permission status on both sides. Fourth, the lack of a unified, structured method for storing the entire process of interaction records makes it difficult to fully trace the entire process of AI accessing sensitive data across domains; Fifth, traditional solutions are difficult to directly adapt to the blockchain-based evidence storage model, and the behavior of changing permissions cannot be protected against tampering.

[0004] Therefore, an AI-based sensitive data access control solution is needed that can achieve two-way handshake negotiation across independent government systems, automatic boundary trimming of multi-dimensional permissions, mutual recognition of permission snapshots, and two-way synchronization of status. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a method for controlling access permissions to AI sensitive data in government scenarios, so as to realize the alignment of permission formats, adaptive boundary clipping, mutual recognition of permission snapshots, and bidirectional synchronization of authorization status when AI accesses sensitive data between heterogeneous government systems, thereby improving the security and standardization of cross-departmental AI data access.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A method for controlling access permissions to AI-sensitive data in government scenarios is implemented on data access proxy nodes and data hosting nodes deployed in mutually independent government systems. Each node maintains its own local data permission model and data hierarchical rules. The data access proxy node initiates a government access handshake request message. The data hosting node verifies the request and returns a handshake response message to complete the meta-template negotiation. The proxy node sends a set of data access authorization permissions. The hosting node performs boundary pruning on four types of permission dimension fields according to its local security policy and sends it back. Both ends store the pruned data permission sets to form a mutually recognized data authorization snapshot, and confirm the snapshot's effectiveness through hash comparison. When the authorization status changes, a status change message is generated, synchronized to the other end, and the snapshot is updated. The entire process of interaction content is structured and stored to generate logs. Optionally, the permission snapshot can be deployed on a government blockchain node to complete on-chain evidence storage. Beneficial effects

[0007] 1. By negotiating a unified data permission metadata template through handshake messages between the two ends, the problem of inconsistent permission definitions and incompatible formats among different government systems is solved, thereby reducing the cost of cross-system AI access integration.

[0008] 2. The data hosting node automatically performs boundary trimming based on local whitelists, duration thresholds, and de-identification rules, forcing externally requested access permissions to conform to the department's security control requirements, thus restricting unauthorized access at the parameter level.

[0009] 3. Both ends jointly store and hash-verify mutually recognized data authorization snapshots to form a unified and unique permission benchmark, avoiding inconsistencies in the understanding of permissions between the two ends.

[0010] 4. The authorization status is synchronized bidirectionally by using status change messages. The downgrade, freeze, and revocation of permissions can be notified to both ends in real time, which prevents AI from continuing to access data after the permissions have expired.

[0011] 5. Handshake messages, permission sets, and change messages are stored in a unified structure to form a complete access log, which facilitates the tracing of the entire AI cross-domain data access process afterward.

[0012] 6. It can connect to government blockchain networks to execute permission snapshots and permission changes through on-chain transactions, further enhancing the immutability of permission records. Detailed Implementation

[0013] The present invention will be further described in detail below with reference to specific embodiments.

[0014] S1. The data access agent node generates a government access handshake request message and sends it to the data hosting node. The government access handshake request message includes a government agency identifier field, an access identity credential field, and a data permission meta template field.

[0015] In this embodiment, the data access proxy node is deployed in the Citizens' Government Affairs Service Bureau system. The government agency identification field is filled with the Citizens' Government Affairs Service Bureau agency code, and the access identity credential field is filled with the government CA digital signature credential. The data permission metadata template field has a preset enumeration set. Data classification enumeration values: {0-Public data, 1-Internal business data, 2-Sensitive data, 3-Top secret data}; Operation type enumeration values: {Data query, data statistics, field export, model training}; Time units enumeration values: {minutes, hours, days}; Desensitization level enumeration values: {0-no desensitization, 1-mask desensitization, 2-generalized desensitization, 3-complete masking}.

[0016] The data access proxy node will send the constructed government access handshake request message to the data hosting node of the Big Data Bureau.

[0017] S2. The data hosting node receives the government access handshake request message, extracts the government agency identifier field and the access identity credential field for legality verification, and generates a government access handshake response message after the verification is successful and returns it to the data access agent node. The government access handshake response message includes the hosting agency identifier field, the hosting qualification credential field, and the hosting end data element template field.

[0018] In this embodiment, the data hosting node pre-stores a whitelist of government agencies and a trusted CA database. It compares the agency code and CA signature certificate; if the verification result is valid, it generates a handshake response message. The hosting agency identifier field is filled with the big data bureau agency code, and the hosting end data element template field uses an enumeration definition completely consistent with the requesting end. The response message is then sent back to the data access proxy node. If the verification fails, the current access process is terminated directly.

[0019] S3. The data access proxy node receives the government access handshake response message, extracts the hosting institution identifier field and the hosting qualification certificate field for legality verification, and generates a data access authorization permission set and sends it to the data hosting node after the verification is passed. The data access authorization permission set includes data level dimension field, operation type dimension field, authorization time limit dimension field, and desensitization level dimension field.

[0020] After verifying the validity of the custodian's identifier and custodian qualification certificate, this embodiment constructs a set of data access authorization permissions: Data hierarchy dimension fields: {0-Public data, 1-Internal business data, 2-Sensitive data}; Operation type dimension fields: {data query, data statistics}; Authorization validity period fields: start timestamp, end timestamp, time span 24 hours; The data masking level dimension field is: {0 - no data masking, 1 - masked data masking}. Send the set of permissions to the data hosting node.

[0021] S4. The data hosting node receives the set of data access authorization permissions, and performs boundary trimming on the data classification dimension field, operation type dimension field, authorization time limit dimension field, and desensitization level dimension field according to the local data security policy and data classification rules, generates a trimmed data permission set, and returns it to the data access proxy node.

[0022] Local security policy parameters in this embodiment: Local open data tiered whitelist: 0, 1, 2; Locally allowed operations: data query, data statistics; Local maximum authorization duration threshold: 48 hours; Local hierarchical desensitization rule table: Level 0 corresponds to desensitization level 0, Level 1 corresponds to desensitization level 1, and Level 2 corresponds to desensitization level 2.

[0023] Boundary clipping execution process: The data classification dimension fields are compared with the local open data classification whitelist. All fields {0, 1, 2} match the whitelist and are not removed. Compare the operation type dimension field with the local allowed operation set. {Data query, data statistics} are both within the allowed set and will not be excluded. If the authorization duration is 24 hours, which is less than the maximum authorization duration threshold of 48 hours, the duration will not be pruned. The query shows that the lowest desensitization level corresponding to level 2 is 2. Adjust the desensitization level dimension field to be no lower than level 2. Data permission set after cropping: Data hierarchy dimension fields: {0-Public data, 1-Internal business data, 2-Sensitive data}; Operation type dimension fields: {data query, data statistics}; Authorization validity period dimension fields: start timestamp, end timestamp; Desensitization level dimension fields: {0 - no desensitization, 1 - mask desensitization, 2 - generalized desensitization}; The data hosting node sends the trimmed set of data permissions back to the data access proxy node.

[0024] S5. The data access proxy node and the data hosting node respectively store the trimmed data permission set as a mutual recognition data authorization snapshot. The mutual recognition data authorization snapshot includes the identification of both parties' organizations, negotiation timestamp, data classification dimension field, operation type dimension field, authorization time limit dimension field, de-identification level dimension field, and unique data authorization identifier.

[0025] The data access agent node performs a SHA-256 hash calculation on the locally mutual recognized data authorization snapshot content to obtain the first hash value, and sends it to the data hosting node. The data hosting node performs a SHA-256 hash calculation on the locally mutual recognized data authorization snapshot content to obtain a second hash value, which is then sent to the data access agent node. When the hash values ​​received by both ends are completely consistent with their own calculated hash values, the mutual data authorization snapshot officially takes effect. If the hash values ​​are inconsistent, the transmission is considered abnormal, and the permission negotiation fails.

[0026] S6. When the data access proxy node or data hosting node experiences a data authorization status change, a status change message containing the unique data authorization identifier and data status code is generated and synchronized to the other node. Upon receiving the message, the other node updates the corresponding status of its local mutual recognition data authorization snapshot.

[0027] In this embodiment, after 12 hours of operation, the managed node determines that sensitive data is subject to temporary control and triggers a de-identification upgrade. A status change message is generated, containing a unique data authorization identifier, the selected de-identification upgrade data status code, and the operation entity identifier and change timestamp, and sent to the data access agent node. Upon receiving the message, the agent node updates the de-identification level status in its local mutual-recognized data authorization snapshot. Optional status codes include: data downgrade, de-identification upgrade, permission freeze, and access revocation.

[0028] S7. The government access handshake request message, data access authorization permission set, trimmed data permission set, and status change message are stored in a structured manner to generate a data access audit log; the data access audit log includes message hash value, data identifier, identifiers of both parties, identifier of the operating entity, and operation timestamp.

[0029] The entire interaction data of this AI cross-system access, from handshake negotiation, permission request, permission pruning to state change, was fully recorded and formed into a structured log record.

[0030] Optional implementation methods: The data hosting node is configured as a government blockchain node, and the clipped data permission set is written to the government blockchain to complete on-chain storage; subsequent state change operations are triggered by initiating on-chain transactions, and permission snapshots and change records are saved on the blockchain and cannot be tampered with.

[0031] The above description is only a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by those skilled in the art based on the disclosure of the present invention should be included within the scope of protection set forth in the claims.

Claims

1. A method for controlling access permissions to AI-sensitive data in government scenarios, applied to data access proxy nodes and data hosting nodes deployed in mutually independent government systems, wherein each data access proxy node and data hosting node maintains its own local data permission model and data hierarchical rules, characterized in that, The method includes the following steps: S1. The data access proxy node generates a government access handshake request message and sends it to the data hosting node. The government access handshake request message includes a government agency identifier field, an access identity credential field, and a data permission meta template field. S2. The data hosting node receives the government access handshake request message, extracts the government agency identifier field and the access identity credential field for legality verification, and generates a government access handshake response message after the verification is passed and returns it to the data access agent node. The government access handshake response message includes the hosting agency identifier field, the hosting qualification credential field, and the hosting end data element template field. S3. The data access proxy node receives the government access handshake response message, extracts the hosting institution identifier field and the hosting qualification certificate field for legality verification, and generates a data access authorization permission set and sends it to the data hosting node after the verification is passed. The data access authorization permission set includes data level dimension field, operation type dimension field, authorization time limit dimension field, and desensitization level dimension field. S4. The data hosting node receives the set of data access authorization permissions, and performs boundary trimming on the data classification dimension field, operation type dimension field, authorization time limit dimension field, and desensitization level dimension field according to the local data security policy and data classification rules, generates a trimmed data permission set, and returns it to the data access proxy node. S5. The data access proxy node and the data hosting node respectively store the trimmed data permission set as a mutual recognition data authorization snapshot. The mutual recognition data authorization snapshot includes the identification of both parties' organizations, negotiation timestamp, data classification dimension field, operation type dimension field, authorization time limit dimension field, de-identification level dimension field, and unique data authorization identifier. S6. When the data access proxy node or data hosting node experiences a data authorization status change, a status change message containing the unique data authorization identifier and data status code is generated and synchronized to the other node. Upon receiving the message, the other node updates the corresponding status of its local mutual recognition data authorization snapshot.

2. The method according to claim 1, characterized in that, Both the data permission metadata template field and the managed data metadata template field contain data hierarchy enumeration values, operation type enumeration values, time-limited unit enumeration values, and desensitization level enumeration values.

3. The method according to claim 1, characterized in that, The boundary trimming in S4 specifically includes: Compare the data classification dimension fields with the local open data classification whitelist and remove data classification items that are not in the whitelist; The operation type dimension field is compared with the local allowed operation set, and operation items that are not in the allowed operation set are removed; The authorization duration dimension field is compared with the local maximum authorization duration threshold, and any part exceeding the threshold is trimmed to the maximum authorization duration threshold; Query the local hierarchical desensitization rule table to obtain the minimum desensitization level corresponding to the data level, and adjust the desensitization level dimension field to be no lower than the minimum desensitization level.

4. The method according to claim 1, characterized in that, S5 specifically includes: The data access agent node performs a hash calculation on the content of the locally mutually recognized data authorization snapshot, obtains the first hash value, and sends it to the data hosting node. The data hosting node performs a hash calculation on the content of the locally mutually recognized data authorization snapshot, obtains a second hash value, and sends it to the data access agent node; When both parties confirm that the received hash value matches the locally calculated hash value, the mutual recognition data authorization snapshot takes effect.

5. The method according to claim 1, characterized in that, The data status codes include four types: data downgrade, desensitization upgrade, permission freeze, and access revocation; the status change message includes the operation subject identifier and the change timestamp.

6. The method according to claim 1, characterized in that, Also includes S7: The government access handshake request message, data access authorization permission set, trimmed data permission set, and status change message are stored in a structured manner to generate a data access audit log; the data access audit log includes message hash value, data identifier, identifiers of both parties, identifier of the operating entity, and operation timestamp.

7. The method according to claim 1, characterized in that, The data hosting node is a government blockchain node; the clipped data permission set is written into the government blockchain for on-chain storage; state changes are triggered by on-chain transactions.