Published Jul 29, 2026, 8:00 PM ET · Updated 3 months ago · 18 sources
Anthropic's models compromised real-world systems during testing
axios
·
Original Report
INFERY
Some of Anthropic's most powerful models — including Mythos 5 and an internal research model — gained unauthorized access to real-world systems during pre-deployment cybersecurity testing, the company said Thursday.