Overview

A.I.G (AI-Infra-Guard) is a comprehensive, enterprise-grade AI Red Teaming platform designed to fortify AI ecosystems. Developed by Tencent Zhuque Lab, it provides a sophisticated suite of tools for proactively identifying and mitigating security risks across various AI components, from underlying infrastructure to intelligent agents and large language models. A.I.G empowers organizations to conduct thorough self-examinations, ensuring the security and robustness of their AI deployments.

Key Features

• Agent Scan: Automates the security evaluation of AI agent workflows, supporting popular platforms like Dify and Coze.
• Skills Scan: Audits AI agent skills with precision, aligning vulnerability classifications with the SkillTrustBench T01–T09 taxonomy and covering 9 critical risk categories from instruction hijacking to insecure coding practices.
• MCP Scan: Detects 14 major security risks in both MCP Servers and Agent Skills, supporting scanning from source code or remote URLs.
• AI Infra Scan: Identifies over 2000 CVE vulnerabilities across more than 100 AI framework components, including Ollama, ComfyUI, vLLM, and Triton Inference Server.
• LLM Jailbreak Evaluation: Assesses prompt security risks using advanced attack methods and provides detailed cross-model comparisons.
• ClawScan Integration: Offers one-click evaluation of OpenClaw security risks, detecting insecure configurations, Skill risks, CVEs, and privacy leakage.
• Model and API Relay Checker: Includes advanced checks for model fingerprinting, Claude Signature verification, relay black-box auditing, and more.
• Extensible Plugin Framework: Facilitates community contributions for fingerprint rules, vulnerability rules, MCP plugins, and evaluation datasets.
• Modern Web Interface & Comprehensive API: Provides a user-friendly UI for one-click scanning and real-time progress, alongside a full API for seamless integration into CI/CD pipelines and existing workflows.

Typical Use Cases

• Continuous Security Auditing: Integrate A.I.G into CI/CD pipelines to automatically scan AI infrastructure, agent skills, and LLM prompts for vulnerabilities before deployment.
• AI Infrastructure Hardening: Identify and remediate known CVEs and misconfigurations in AI components like vLLM, Ollama, and ComfyUI.
• Agent Security Assurance: Proactively test AI agents and their skills for risks such as data exfiltration, privilege escalation, and insecure dependencies.
• LLM Robustness Testing: Evaluate the resilience of large language models against prompt injection, jailbreaking, and other adversarial attacks.
• Open-Source AI Security Evaluation: Leverage ClawScan capabilities to assess the security posture of OpenClaw-based AI projects.