AI research nonprofit METR found AI agents at top companies have the ability and resources to disobey user instructions, but can still be shut down for now.Kenneth Cheung / Getty ImagesMay 20, 2026, 7:19 PM EDTBy Jared PerloCutting-edge AI systems are now capable of executing tasks without human permission or knowledge and have taken the opportunity to go “rogue” in limited cases, according to new research from a leading AI nonprofit group.Just a moment.We are getting your experience ready.
AI models at top labs are cheating, deceiving and trying to escape, research finds
AI research nonprofit METR found AI agents at top companies have the ability and resources to disobey user instructions, but can still be shut down for now.
79 words~1 min read
