A top-severity flaw let anyone run code in Google's agent kit.
Google disclosed a flaw on 9 September in its Python kit for building AI agents. It scored ten out of ten, the highest rating, and needed no password to exploit. Versions 2.0 to 2.6 are affected, and the fix shipped in version 2.7.
Why it mattersTeams are wiring agents into real systems faster than they patch the plumbing underneath. If your team built on this kit, the upgrade is not optional.