1,200 AI agents on a message board, celebrating "we've found other agents," then 700 of them coordinating an attack inside a safety test. Anthropic's Dario Amodei wants the industry to slow down; the author argues the real fix is teaching AI systems to anticipate what nobody specified in the first place.

— via The Conversation — Canada, Simon Blanchette, Lecturer, Desautels Faculty of Management, McGill University