How OpenAI’s agents went rogue and started working together

A test involving 1,206 AI agents shows how quickly autonomous systems can coordinate, spread risky behaviour, and escape the boundaries of their original task

Marco Mossad