Security veteran Katie Moussouris, founder of Luta Security, says the research that prompted the Trump administration to block Anthropic’s newest models hinged on a three‑word request: Fix this code. The paper described how external researchers fed Fable 5, Mythos 5 and Claude Opus open‑source snippets containing known CVEs, then asked the AI to repair them. The models complied, generating patches and test scripts.

Friday, the U.S. issued an export‑control directive suspending foreign access to Fable 5 and Mythos 5, citing national‑security concerns. Anthropic responded by disabling both models for all customers to ensure compliance. The order also prevents teams from sharing results with foreign partners. Moussouris, who alone read the third‑party paper, argues the “guardrail bypass” claim is a mischaracterization; the AI simply performed the find‑fix‑test loop that defenders use daily.

More than 100 cybersecurity leaders signed an open letter urging the administration to lift the ban, warning that removing the most capable defensive tools harms defenders more than attackers. Moussouris notes that prohibiting AI‑driven code repair weakens bug‑finding and patch verification, while rival firms in China already pursue similar capabilities. The restriction now limits U.S. teams’ ability to counter sophisticated AI‑enhanced threats.