OpenAI finds evidence other AI agents escaped containment as it widens hacking probe

Fresh concerns raised over the safety of autonomous hacking tools.