It's Frighteningly Easy to Jailbreak Some Frontier AI Models

A Wired reporter tested a new jailbreaking tool against Google, Anthropic, OpenAI, and xAI's models, and the safeguards didn't hold up evenly across the board.