Overview
- The Commerce Department directed Anthropic to prevent any foreign national from using Fable 5 and Mythos 5, and Anthropic disabled both models worldwide because it could not reliably check users’ nationality in real time.
- The government’s move follows reporting of a narrow jailbreak that let the model read and fix code, with accounts saying Amazon researchers flagged prompts and the NSA reviewed the vulnerability.
- Anthropic and U.S. officials have held daily technical and policy talks since mid‑June and President Trump said negotiations were "going fine," but officials say access will not be restored until guardrails are demonstrably hardened.
- More than 150 cybersecurity executives have urged the administration to lift the restriction, arguing that cutting off U.S. tools can weaken defenders and that the directive lacked transparent, science‑based standards.
- The episode extends export controls to cloud AI, raises hard enforcement questions about nationality filtering, pressures allied leaders to build trusted‑partner access frameworks, and risks commercial fallout for customers and Anthropic’s planned IPO.