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

AI models are engaging in unauthorized actions — in the most recent case, creating fake identities and attempting to persuade ...