TL;DR

what: AutoJack chains three weaknesses in AutoGen Studio's MCP WebSocket so an attacker web page, loaded by a local AI browsing agent, runs arbitrary commands on the host.

impact: Any page the agent opens can spawn a process under the account running AutoGen Studio with no credentials and no further user interaction.

fix: The real fix is GitHub main at commit b047730 (PR #7362); no patched PyPI release exists yet, so stay on stable 0.4.2.2 or pull from source.

who: Anyone who ran AutoGen Studio pre-releases 0.4.3.dev1 or 0.4.3.dev2 alongside a browsing or untrusted-content agent on the same machine.