Fooling AI with poetry: Why systems’ safety controls are not very effective

Just over three years after the debut of ChatGPT, tricking artificial intelligence into bad behaviour is almost a trivial exercise