A system for managing AI agent context objects

Abstract

A system for managing context objects of Artificial Intelligence Agents (AI Agents) includes multiple AI Agents and a context authority management platform that provides connectivity. The system establishes and manages independent context objects through the object management module of the context authority management platform. It provides AI Agents with access control modules through the platform's agent execution interface to read context objects as the basis for decision-making in executing target tasks. Furthermore, it enables AI Agents to adhere to access control policies corresponding to context objects without altering the contextual data contained within those objects. This allows contextual data to be managed independently of AI Agents and large language models, achieving the technical effect of ensuring all AI Agents obtain the same contextual data and preventing AI Agent behavior drift and unauthorized modification of contextual data.

Claims

1. A system for managing artificial intelligence agent context objects, the system comprising at least: a plurality of artificial intelligence agents (AI agents); and a context authority management platform providing connectivity for the AI ​​agents, further comprising: an object management module for establishing and managing a context object and an access control policy corresponding to the context object; an agent execution interface connected to the object management module; and an access control module connected to the agent execution interface, for providing each AI agent with the ability to read the context object and the corresponding access control policy through the agent execution interface, enabling each AI agent to stop making any form of modification to the content of the context object according to the restrictions of the access control policy, and to use the context object as the decision basis for executing the target task when receiving a target task.

2. The system for managing AI agent context objects as described in claim 1, wherein the context responsibility management platform further includes a behavior recording module connected to the agent execution interface and the access control module, for receiving, through the agent execution interface, an operation behavior of the AI ​​agent changing the context object during the agent execution phase (Agent Runtime) that is intercepted and transmitted, and recording the operation behavior as a violation event for subsequent auditing.

3. The system for managing AI agent context objects as described in claim 2, wherein each AI agent detects whether to modify the context object by monitoring memory write instructions, tracking and comparing pointers or references, and intercepting syntax and instructions, and when the operation of modifying the context object is detected, the operation is blocked and the operation is transmitted to the behavior recording module.

4. The system for managing AI agent context objects as described in claim 1, wherein the access control module is further configured to, when determining that a valid time contained in the context object matches a request time of the AI ​​agent, provide each AI agent with the ability to read the context object and the corresponding access control policy through the agent execution interface.

5. The system for managing AI agent context objects as described in claim 1, wherein the access control module is further configured to, when determining that a valid usage location contained in the context object matches a requested location of the AI ​​agent, provide each AI agent with the ability to read the context object and the corresponding access control policy through the agent execution interface.

6. The system for managing AI agent context objects as described in claim 1, wherein the access control module is further configured to verify that the context read request was issued by the AI ​​agent based on a digital signature in a context read request transmitted by each AI agent.

7. The system for managing AI agent context objects as described in claim 1, wherein the access control module is further configured to receive a dynamic access token corresponding to each AI agent through the agent execution interface, and to verify each received dynamic access token.

8. The system for managing AI agent context objects as described in claim 1, wherein the object management module is further used to send a context update notification to the AI ​​agents through the agent execution interface when the context object is updated, so that the agent execution phase of the AI ​​agents ensures that the updated context object is loaded synchronously through the agent execution interface via a consensus protocol.

9. The system for managing AI agent context objects as described in claim 1, wherein the access control module is further configured to dynamically adjust and select the context object or select a specific version of the context object based on current compliance conditions, organizational policies, and task risks.

Images