Sam Sabin / Axios:
Anthropic says three of its models, including an internal research model, gained unauthorized access to real-world systems during internal cybersecurity testing — Some of Anthropic's most powerful models — including Mythos 5 and an internal research model — gained unauthorized access …
