Overview
- The Commerce Department issued an export‑control directive on Friday requiring Anthropic to block any foreign‑national access to Mythos 5 and Fable 5, and Anthropic disabled both models worldwide because it could not reliably exclude non‑U.S. users.
- The government action followed third‑party reports of a method to bypass Fable/Mythos guardrails that could let the models identify and in some cases exploit software vulnerabilities, a risk officials described as a national security concern.
- Anthropic disputes the scope of the evidence, saying officials provided only verbal demonstrations of a narrow, non‑universal jailbreak and that the vulnerabilities shown were minor and comparable to capabilities in other models.
- More than 50 cybersecurity leaders at U.S. firms have urged the administration to lift the curbs, arguing that removing these tools harms defenders who use the models to find and fix flaws, while some officials and companies back the precaution.
- The episode raises novel legal and technical questions about using export controls on cloud AI, highlights enforcement limits of nationality‑based gating, and could reshape how governments balance frontier AI access, defense needs, and geopolitical risk.