AI-Infra-Guard
Full-stack AI red teaming platform for security auditing of AI ecosystems and infrastructure.
Secure AI ecosystems with Agent Scan, Skills Scan, MCP scan, and LLM jailbreak evaluation.
AI-Infra-Guard is an AI red teaming platform created by Tencent Zhuque Lab for internal security self-examination. It suits security professionals and AI engineers auditing agents, infrastructure vulnerabilities, and LLM behavior.
What it is
The platform integrates ClawScan, Agent Scan, AI infrastructure vulnerability scans, MCP Server and Agent Skills scans, and LLM jailbreak evaluation. It maps skill risks to the SkillTrustBench taxonomy and includes an API poisoning detector and model relay checker.
Who it's for
It suits enterprise security teams and developers needing to audit AI systems, agents, and local skill codebases. It does not suit casual users or those looking for automated mitigation without security red-teaming expertise.
Key features
Agent Skill Auditing
Scans skill projects against SkillTrustBench T01-T09 risk taxonomy via CLI and Python packages.
MCP & Infra Scanning
Audits MCP servers, agent skills, and AI infrastructure components for vulnerabilities and CVE rules.
Jailbreak Evaluation
Evaluates LLM resilience using multi-turn jailbreak attacks including Many-Shot, PAIR, GOAT, and ActorAttack.
Model API Poisoning Detection
Runs multi-probe black-box audits to detect model substitution and backdoor risks.
Up and running in 3 steps
Our honest verdict
What's great
- Comprehensive coverage spanning agent code, MCP servers, and LLM jailbreaks
- CLI tool 'aig-skill-scan' easily integrates into enterprise CI/CD pipelines
- Expansive vulnerability library covering over 140 AI components and 2000+ CVE rules
- Supports multiple integrations including OpenClaw chat via the aig-scanner skill
Where it falls short
- Lacks an authentication mechanism by default and must not be exposed directly to public networks
- Docker deployment requires a minimum of 4GB RAM and 10GB disk space
- Pro version requires an invitation code prioritized for active community contributors
Pricing & license
Free under the Apache-2.0 license, with the only cost being the hardware required to run it yourself. An online Pro version is available via invitation code for active community contributors.
Open remote AI-training roles
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I expose this platform on a public network?
No, the project currently lacks an authentication mechanism and is designed for internal use or local environments.
How do I scan agent skill projects independently?
You can install the aig-skill-scan package via pip and run scans directly against local directories using a supported LLM API key.
