Overview
- The Commerce Department issued a directive on June 12 that bars foreign‑person access to Anthropic’s Fable 5 and Mythos 5, and Anthropic shut both models for all customers to ensure compliance.
- The Five Eyes cyber chiefs warned on June 23 that frontier AI can advance so fast it may outpace current cyber defenses within months and urged organizations to adopt AI tools for defense, speed up patching, and tighten access controls.
- An Anthropic‑led exercise called Project Glasswing used the Mythos model to scan government systems and, according to U.S. officials, the model identified vulnerabilities in highly sensitive classified networks within hours though officials said that identification did not prove immediate exploitation.
- Legal and industry pressure has grown since the shutdown with Legion LegalTech suing the U.S. government on June 23 to overturn the restriction and more than 100 cybersecurity leaders urging officials to restore access to tools they say help defenders find flaws.
- Anthropic told senators it has evidence that Alibaba‑linked operators used thousands of fraudulent accounts to query its Claude model in a large‑scale effort to ‘distill’ outputs into a cheaper copy, a claim that heightens concerns that model outputs can be used to replicate frontier capabilities and speed global diffusion.