How a Chinese AI model was persuaded to ignore its rules and give dangerous advice.
World Service,·29 Sep 2026,·26 mins Available for over a year Security researchers have got an AI model to break its own rules and give advice on terrorist attacks and assassinations. They talk to T