一种智能体动态编排与自适应授权边界控制方法及系统
By employing a dynamic orchestration and adaptive authorization boundary control method for intelligent agents, the security level is calculated in real time, and a lightweight BERT-base model is used for content-level sensitivity re-judgment. This solves the problems of implicit unauthorized access leakage and business continuity in large language models, realizes the reliable transmission and dynamic adjustment of security level labels, and improves the task completion rate.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANJIN THINKING CHAIN ARTIFICIAL INTELLIGENCE CO LTD
- Filing Date
- 2026-05-12
- Publication Date
- 2026-07-17
AI Technical Summary
In large language model intelligent agent scenarios, traditional static permission models cannot cover the drift of dynamic authorization boundaries, leading to implicit unauthorized disclosure and business continuity issues. Moreover, existing security systems usually terminate the task directly after detecting risks, sacrificing business continuity.
By employing a dynamic agent orchestration and adaptive authorization boundary control method, a directed acyclic graph is dynamically generated to calculate the security level in real time. A lightweight BERT-base model is used for content-level sensitivity re-judgment. Combined with a structured security label envelope and chain signature mechanism, the trusted transmission and dynamic adjustment of the security level are realized.
It effectively reduces the risk of implicit unauthorized access and ensures the reliable transmission of security level labels, improves the completion rate of complex tasks and business continuity, adapts to dynamically generated task DAGs, and solves the problem of static permission models failing in AI dynamic orchestration.
Smart Images
Figure CN122174215B_ABST