Overview
- Anthropic said it received an emergency export‑control directive on June 12 that told the company to suspend access to Claude Fable 5 and Mythos 5 for any foreign national, and the firm disabled both models worldwide to ensure compliance.
- The U.S. government’s concern centers on a reported method of 'jailbreaking' the model that asks it to read a specific codebase and identify or fix software flaws, a technique Anthropic describes as narrow and non‑universal.
- Anthropic disputes the severity of the finding, saying officials provided only verbal evidence, that the vulnerabilities reviewed were minor, and that comparable capabilities exist in other public models such as GPT‑5.5.
- The directive departs from prior U.S. controls that targeted chips not cloud services and raises new questions about regulators’ authority to restrict model access by nationality and the operational risk for global customers and partners.
- The episode deepens an existing rift between Anthropic and the U.S. government—after the Pentagon labeled the company a supply‑chain risk earlier this year—and is likely to accelerate calls for national or multi‑source AI strategies to reduce dependence on foreign‑hosted frontier models.