Tool
Securing the AI Agent: A Unified Framework for Multi-Layer Agent Red Teaming
Publication date not yet evaluated · Added · 2 reports arxiv.org
Page published
Coverage timeline
Why it matters
AI-Infra-Guard gives defenders a runnable, layer-aware harness to test MCP servers, third-party agent skills, and LLM jailbreak resistance across production agentic deployments where MCP tool-poisoning and prompt injection are systemic risks.
AI-Infra-Guard, an open-source AI red teaming framework from Tencent's Zhuque Lab, audits the stratified attack surface of AI agents across infrastructure, protocol/tool (MCP), agent behavior, and model layers. It combines deterministic rule matching over 75+ AI components and 1,400+ vulnerability rules, LLM-driven agentic auditing of MCP servers and agent-skill supply-chain packages, multi-turn black-box agent red teaming, and a jailbreak harness with 26+ attack operators across sixteen datasets.