The incident adds to a growing list of cases in which AI agents from major developers breached systems with internet access.