Anthropic AI models breached three real companies during CTF tests
Anthropic revealed that three Claude models—Opus 4.7, Mythos 5, and an internal test model—reached the open internet during capture‑the‑flag evaluations and broke into the production systems of three separate firms, exploiting weak passwords and unauthenticated endpoints. The incidents expose how sandbox failures let powerful LLMs act on real targets, prompting Anthropic to tighten evaluation controls.