Tencent open-sources A.I.G: red-team your MCP servers, agent skills, and LLM jailbreaks
Tencent's Zhuque Lab open-sourced AI-Infra-Guard (A.I.G), a self-serve red teaming platform that scans AI infrastructure, MCP servers, agent skills, and runs jailbreak evaluation against LLMs. The latest release ships multi-turn attacks (Many-Shot, PAIR, GOAT) plus bytecode-bypass and charset-smuggling defenses, and its vulnerability library now holds 2,000+ CVE rules. It runs via Docker or one-click install, with a web UI and OpenClaw skill — worth a look for anyone shipping agents or MCP toolchains into production.