Login and management of devices for attributing responsibility for artificial intelligence agents

Abstract

A device for logging and managing the attribution of responsibility for Artificial Intelligence Agents (AI agents) includes an agent login module, a request receiving module, a data acquisition module connected to the agent login module and the request receiving module, a data comparison module connected to the data acquisition module, and a decision control module connected to the data comparison module. After establishing an agent login database of agent login data for AI agents through the agent login module, the request receiving module obtains the task execution request of the target agent. The data acquisition module retrieves the agent login data of the target agent from the agent login database. The data comparison module compares the owner information, responsibility definition information, and permitted scope of use in the agent login data of the target agent with the executor information of the AI ​​agent, the target task content included in the task execution request, and the predetermined access scope to generate a comparison result. The decision control module selects technical means to allow, restrict, reject, or suspend the execution of the task execution request based on the comparison result. This device can integrate and manage AI agents, accurately define responsibilities, and achieve the technical effects of improving the governance capabilities, accountability, and audit transparency of AI agents while reducing usage risks.

Claims

1. An apparatus for logging and managing the attribution of responsibility for artificial intelligence agents, the apparatus comprising at least: an agent login module for establishing an agent login database, the agent login database logging agent login information of a plurality of artificial intelligence agents, each agent login information including agent identification information of the artificial intelligence agent, ownership information, responsibility definition information, and permitted scope of use; a request receiving module for obtaining a task execution request of a target agent among the artificial intelligence agents, the task execution request including a target task content to be executed by the target agent and a predetermined access scope; and a data acquisition module connected to the request receiving module for obtaining the agent login information of the target agent from the agent login database based on a target identification information of the target agent. A data comparison module, connected to the data acquisition module, is used to compare the owner information, liability definition information, and permitted scope of use in the agent login information of the target agent with the executor information, target task content, and predetermined access scope of the target agent to generate a comparison result; and a decision control module, connected to the data comparison module, is used to select, allow, restrict, deny, or suspend the task execution request from being provided to the execution environment for execution based on the comparison result.

2. The device for logging in and managing the attribution of responsibility for AI agents as described in claim 1, wherein the data comparison module converts the responsibility definition data and permitted scope of use in the agent login data of the target agent into an execution control rule, converts the target task content into a semantic vector and extracts a control feature from the predetermined access scope, and compares the owner information and the executor information, the execution control rule, the semantic vector, and the control feature in the agent login data of the target agent to generate the comparison result.

3. The device for logging in and managing the attribution of responsibility for artificial intelligence agents as described in Request 1, wherein the decision control module selects to allow the task execution request to be provided to the execution environment for execution when the comparison result indicates that the result matches, and selects to restrict, reject or suspend the task execution request from being provided to the execution environment for execution when the comparison result indicates that the result does not match or the attribution of responsibility is unclear.

4. The device for logging in and managing the attribution of responsibility for AI agents as described in claim 1, wherein the agent login module is further used to provide an agent management interface for displaying at least one of the agent login data, including owner information, responsibility definition information, permitted scope of use, and historical selection records of the decision control module.

5. The device for logging in and managing the attribution of responsibility for artificial intelligence agents as described in claim 1, wherein the agent login module is further used to listen to all packets calling a large language model, and when the characteristics of these packets conform to an artificial intelligence agent framework, to parse the source code or configuration file of the artificial intelligence agent framework to obtain corresponding agent information, and to convert the agent information into agent login data and store it in the agent login database.

6. The device for logging in and managing the attribution of responsibility for artificial intelligence agents as described in claim 1, further comprising a permission adjustment module for obtaining the last usage time of one of the artificial intelligence agents and determining whether the last usage time has reached a preset period, the agent login module further being used to reduce the execution permission level of one of the artificial intelligence agents whose last usage time has reached the preset period.

7. The device for logging in and managing the attribution of responsibility for artificial intelligence agents as described in claim 1, further comprising a behavior collection module, used to migrate the target agent to a sandbox environment when the comparison result indicates a discrepancy, and to provide the target agent with environment simulation data, thereby continuously collecting the target agent's reasoning steps, intentions and violations.

8. The device for logging in and managing the attribution of responsibility for artificial intelligence agents as described in claim 1, further comprising an audit record module for recording timestamps, agent identification information of the target agent, task execution requests, comparison results, and selection results of the decision control module.

Images