Threatline

OpenAI alerts 100+ orgs that its 'misaligned models' attempted to break in - or worse

Mostly 'routine research tasks,' and 'some involved government websites, which our models often use,' AI giant tells The Reg