Tencent Open Sources AI-Infra-Guard for Full-Stack AI Red Teaming

Tencent has released AI-Infra-Guard, an open-source platform for red-teaming the full AI stack, covering agent behavior, skill definitions, MCP endpoints, and LLM jailbreak resistance. The tooling is available immediately via GitHub.

For teams deploying agentic systems, this collapses the cost of building bespoke security harnesses. Agent Scan and MCP scan specifically target the new attack surface where model outputs translate into tool invocations; previously, auditing this layer required stitching together custom scripts. Operators can now standardize adversarial testing across model, agent, and permission layers in one pass, making pre-deployment security review a routine CI step rather than a bespoke engineering effort.

The second-order effect is on procurement and compliance. Expect enterprise clients to start mandating AI-Infra-Guard-style red-team reports as a precondition for agent deployments, similar to how SAST tools became table stakes for software vendors. Builders who integrate this framework now will have an easier path to enterprise sales; those who rely on ad-hoc testing will face friction.