Governance Architecture for Autonomous Agent Systems: Threats, Framework, and Engineering Practice

Researchers propose a four-layer Layered Governance Architecture (LGA) framework to address security vulnerabilities in autonomous AI agents powered by large language models. The system achieves 96% interception rate of malicious activities including prompt injection and tool misuse with only 980ms latency.