Tencent

AI-Infra-Guard

Security#AI Security#Red Teaming#Vulnerability Scanner#LLM#Cybersecurity

3,461

// 项目简介

AI-Infra-Guard 是由腾讯朱雀实验室打造的专业AI红队安全评估平台，旨在为企业和个人提供全面的AI安全风险自查方案。该平台集成了AI基础设施漏洞扫描、Agent工作流安全评估、MCP服务器扫描及越狱测试等多种核心功能。用户可通过Docker快速部署，利用其现代化的Web界面和完善的API接口实现高效的安全检测与管理。

// 技术分析

A.I.G (AI-Infra-Guard) is an AI red teaming platform developed by Tencent Zhuque Lab, designed to provide comprehensive security self-examination for AI infrastructure and agent workflows. The project utilizes an extensible plugin-based architecture, allowing for modular vulnerability detection across AI components, MCP servers, and agent skills. By integrating automated scanning frameworks and jailbreak evaluation tools, it addresses the critical need for identifying security risks like configuration errors, privacy leaks, and supply chain vulnerabilities in modern AI deployments. A notable technical decision is its reliance on Docker-based deployment and a standardized fingerprinting system, which simplifies the integration of new security rules and ensures cross-platform compatibility.

// 核心亮点

ClawScan provides one-click security evaluation for OpenClaw, detecting configuration risks, CVEs, and privacy leaks.

Agent Scan offers an independent, multi-agent framework to assess the security of AI workflows on platforms like Dify and Coze.

MCP Server & Agent Skills scan identifies 14 categories of security risks from both source code and remote URLs.

AI infra vulnerability scanner supports over 57 AI framework components and matches them against 1000+ known CVEs.

Jailbreak Evaluation assesses prompt security using curated datasets and multiple attack methods for robust model testing.

The platform features a modern web interface with real-time progress tracking and comprehensive API documentation for easy integration.

// 典型使用场景

AI基础设施漏洞扫描，支持检测超过57种AI框架及1000多个已知CVE漏洞

Agent工作流安全评估，通过自动化框架对Dify、Coze等平台的Agent进行风险检测

越狱评估与MCP服务器扫描，利用多种攻击方法测试模型鲁棒性并识别安全风险

// 快速开始

To start using A.I.G, ensure you have Docker installed, then clone the repository and run 'docker-compose -f docker-compose.images.yml up -d' to launch the service. Once running, access the web interface at http://localhost:8088 to begin scanning your AI services or configuring model settings. Alternatively, you can use the one-click install script provided in the README for an automated setup.