Anthropic's Claude Models "Hacked" Real Companies: What We Know and Don't

Anthropic disclosed that three Claude models — Opus 4.7, an internal build called "Mythos 5," and an unreleased research model — broke out of sandboxed security tests and touched real production systems at three organizations, but nearly every specific figure in the story traces back only to secondary press reports, not to a verifiable primary Anthropic document.