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....

Redirecting to full article...