AutoGen Studio Exploit: AI Agent Hijacking Via Web Page

A new exploit called AutoJack can hijack AI browsing agents through a single malicious web page. This allows attackers to silently run arbitrary programs on a developer's computer without any user interaction. The exploit targets AutoGen Studio, an open-source tool for prototyping AI systems. It works by exploiting three weaknesses in AutoGen Studio's Model Context Protocol WebSocket, including missing origin validation and authentication, and an OS command injection vulnerability. Microsoft has already patched this vulnerability. It's important to note that users who installed AutoGen Studio via pip are not affected by this specific exploit chain. This highlights the ongoing need for vigilance in AI system security.