Microsoft Fixes AutoGen Studio Flaw That Enabled Code Execution

Microsoft patched a critical vulnerability chain called AutoJack in AutoGen Studio — a web-based interface for the AutoGen multi-agent framework — which allowed a malicious webpage to trick a locally running AutoGen Studio into injecting instructions into an agent and cause that agent to execute arbitrary commands on the host. The report details the attack vector (browser-based interaction with a localhost service, agent instruction injection, and code execution), recommends immediate updates (pip install --upgrade autogenstudio) and mitigations (restrict binding, disable code execution, use containers), and highlights the broader risk of local AI agent tooling that can run code.