Device that injects corresponding state contexts into the job roles of artificial intelligence agents

Abstract

A device for injecting corresponding state contexts based on the role of an Artificial Intelligence Agent (AI Agent) includes a mapping management module, an information acquisition module, a context maintenance module, a context selection module connected to the information acquisition module and the context maintenance module, and a context control module connected to the mapping management module and the context selection module. The device obtains the AI ​​Agent's role information and task requirements through the information acquisition module. The context selection module selects a subset of state contexts from the AI ​​Agent's state context set recorded by the context maintenance module based on the role information and task requirements. The context control module determines whether to change the subset of state contexts to be injected into the AI ​​Agent based on whether the role information and task requirements match the context access range defined by the mapping management module corresponding to the role information. This technical means can effectively limit the state contexts that the AI ​​Agent can obtain, achieving the technical effect of avoiding the risk of excessive information exposure and implementing the minimum necessary information disclosure.

Claims

1. An apparatus for injecting corresponding state contexts according to the job roles of an artificial intelligence agent, the apparatus comprising at least: a mapping management module for defining the context access range corresponding to a plurality of job roles; an information acquisition module for acquiring job role information and a task requirement of an artificial intelligence agent (AI Agent), wherein the job role information corresponds to one of the job roles; and a context maintenance module for recording a set of state contexts of the artificial intelligence agent. A scenario selection module, connected to the information acquisition module and the scenario maintenance module, is used to select a subset of state scenarios from the set of state scenarios based on the job role information and the task requirements; and a scenario control module, connected to the mapping management module and the scenario selection module, is used to directly inject the subset of state scenarios into the AI ​​agent based on a judgment result of whether the job role information and the task requirements are consistent with the access scope of these scenarios, or to restrict or adjust the subset of state scenarios and then inject the restricted or adjusted subset of state scenarios into the AI ​​agent.

2. The device for injecting corresponding state contexts based on the job role of an AI agent as described in claim 1, wherein the context selection module establishes a knowledge graph based on the set of state contexts, and takes the node in the knowledge graph associated with the task requirement as the starting point, and obtains a knowledge subgraph from the knowledge graph based on the permissions of the job role information, and uses the knowledge subgraph as the subset of state contexts, or chunks the set of state contexts to generate multiple data blocks, adds labels to each data block and converts it into an embedding vector, and retrieves the embedding vectors that match the permissions of the job role information and the embedding vectors that match the task requirement to generate the subset of state contexts, or calculates the semantic relevance of each state context in the set of state contexts to the task requirement, obtains each state context that reaches a threshold value in semantic relevance, and deletes those state contexts that do not match the job role information, thereby generating the subset of contexts.

3. The device for injecting corresponding state contexts into the job role of an AI agent as described in claim 1, wherein the context control module converts the job role information, the task requirements, and the context access range into input data, and compares the input data with an access control matrix composed of job role, task type, and context label to generate the judgment result, or establishes a semantic correspondence between the job role information and the task requirements through description logic, and determines whether the semantic correspondence conforms to the context access range through an inference engine to generate the judgment result, or determines the context relevance, role suitability, and sensitivity through a large language model (LM) to generate the judgment result.

4. The device for injecting corresponding state contexts based on the job role of an AI agent as described in Request 1, wherein the context selection module is further used to select a subset of the state contexts based on permissions, policies, and workflow status.

5. The device for injecting corresponding state contexts based on the job role of an AI agent as described in claim 1, wherein the context selection module is further used to blur the sensitive data based on the job role information and the task requirements when it determines that the subset of state contexts contains sensitive data.

Images