TencentA.I.G (AI-Infra-Guard) is an AI red-teaming platform from Tencent Zhuque Lab for checking AI systems for security risks. It combines several scanners in one web interface: an OpenClaw security scan, an agent workflow scanner, an AI infrastructure vulnerability scanner, a scanner for MCP servers and Agent Skills, jailbreak evaluation, and a checker for model APIs and relays.
The scanners cover different layers. The infrastructure scan fingerprints more than 100 AI framework components, including Ollama, ComfyUI, vLLM, n8n, and Triton Inference Server, and matches them against more than 2000 known CVEs. The MCP and skill scan detects 14 categories of risk from source code or remote URLs. Agent Scan is a multi-agent framework that probes agent workflows on platforms such as Dify and Coze. Jailbreak evaluation runs curated datasets and multi-turn attacks (Many-Shot, PAIR, GOAT, ActorAttack) against a target model and compares results across models. The relay checker fingerprints models and audits API relays for model substitution and backdoor risks.
A standalone skill scanner, aig-skill-scan, audits Agent Skills in CI pipelines against the nine-category SkillTrustBench taxonomy. Its benchmark results with different models:
| Model | F1 | Precision | Recall | False positive rate |
|---|---|---|---|---|
| Claude Opus 4.6 | 0.9848 | 0.9725 | 0.9974 | 0.0663 |
| GLM 5.1 | 0.9836 | 0.9701 | 0.9974 | 0.0723 |
| Gemini 3.5 Flash | 0.9792 | 0.9947 | 0.9641 | 0.0120 |
| Kimi 2.6 | 0.9780 | 0.9895 | 0.9667 | 0.0241 |
| DeepSeek v4 Flash | 0.9740 | 0.9868 | 0.9615 | 0.0301 |
The platform is self-hosted with Docker and is intended for internal use, since it has no authentication layer. OpenClaw users can call it from chat through a skill, and a hosted Pro version is available by invitation.
Features
- ClawScan: one-click evaluation of OpenClaw installs for insecure configuration, risky skills, CVEs, and privacy leakage
- Agent Scan: a multi-agent scanner for agent workflows on platforms such as Dify and Coze, with OWASP-aligned checks
- MCP and skill scan: detects 14 categories of risk in MCP servers and Agent Skills from source code or remote URLs
- AI infra vulnerability scan: fingerprints 100+ AI components and checks them against 2000+ known CVEs
- Jailbreak evaluation: curated prompt datasets and multi-turn attacks with cross-model comparison
- Model and relay checker: model fingerprinting, Claude signature verification, and black-box audits for model substitution
- Skill scanner CLI: aig-skill-scan audits skills in CI against the SkillTrustBench T01 to T09 taxonomy
- OpenClaw skill: an aig-scanner skill lets OpenClaw chat drive scans against a running A.I.G service
