V3.5 Preview has been released

[rsinding]

AI-Infra-Guard has been updated to V3.5 preview version. Due to significant updates and refactoring in V3.5, the current version may contain some bugs or instability. Therefore, a preview version is being released first. We welcome everyone to try it out, provide feedback, contribute code, and participate in the project's development. The official V3.5 version will be released later.

The main updates include:

1. Vulnerability Database: Added 100+ CVEs for AI components, supporting the latest React2Shell (CVE-2025-55182) vulnerability detection, affecting multiple well-known AI frameworks such as Dify, NextChat, and LobeChat;
2. MCP Detection: Reconstructed the MCP Scan Agent framework, improving detection speed by over 30%;
3. Jailbreak Assessment: Added 10+ code/behavior control jailbreak attack operators (such as A1Z26, AffineCipher, AsciiSmuggling, etc.), providing stronger jailbreak generalization capabilities;
4. AI Infra Scanning: Added agent + Chromium dynamic screenshot detection for unauthorized access and data leakage risks in AI infrastructure;
5. Fixed and optimized numerous bugs and details, and improved documentation;
   (See the project Release page for details: https://github.com/Tencent/AI-Infra-Guard/releases)