The incident – involving Anthropic's Mythos 5 – was uncovered during testing by the UK's AI Security Institute. A powerful ...