Overview
- On June 12 the Commerce Department ordered Anthropic to block foreign‑person access to its Fable 5 and Mythos 5 models, and Anthropic disabled both models worldwide to ensure compliance.
- A U.S. official said cooperative Project Glasswing testing showed Mythos identified vulnerabilities in highly sensitive classified systems within hours, while officials clarified the tests did not show the model necessarily exploited those flaws in that time.
- More than 100 cybersecurity experts and industry leaders warned that removing access will hurt defenders who use advanced models to find and patch flaws, and rival releases such as OpenAI’s GPT‑5.5‑Cyber have raised questions about consistency in the government response.
- Litigation and negotiation have followed the shutdown: Legion LegalTech sued the Commerce Department on June 23 seeking to overturn the directive and halt its enforcement, and Anthropic is holding daily technical talks with U.S. agencies.
- Intelligence chiefs in the Five Eyes alliance said frontier AI will reshape offensive and defensive cyber risk in months, a warning that could push governments to expand export controls, create trusted‑partner review processes, and accelerate development of non‑U.S. or open models.