Anthropic Pulled Models Over Behavior Experts Call
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- The U.S. Commerce Department sent Anthropic a letter on Friday invoking an obscure export control directive that banned non-Americans — including Anthropic's own employees — from accessing Fable 5 and Mythos 5, citing an unspecified national security concern that the letter did not detail.
- Anthropic shut down both models for all customers to ensure compliance, and the unpublished letter's vague reasoning has fueled speculation that the move was retaliatory rather than technical.
- The Wall Street Journal reports the guardrail-bypass paper was authored by security researchers at Amazon, and Anthropic shared a private copy with cybersecurity veteran Katie Moussouris, founder of Luta Security, for her assessment.
- Moussouris said the described behavior "should never have triggered an export control" — the distinction is between asking a model to "review code for security issues" versus "fix this code," which produce nearly identical results, and any attempt to fix it would "only weaken the model for defense."
- Moussouris and dozens of top security researchers called on the Trump administration to revoke the order, calling the removal of advanced cybersecurity capabilities from U.S. network defenders "dangerous."
- Axios reported that "personality differences" between Anthropic and the Trump administration — not a technical issue with the AI products — drove the export directive, a characterization the administration has not confirmed.
- Justin Hendrix of Tech Policy Press warned the move is "likely to raise alarms in foreign capitals about the reliability of American AI for critical applications" and that the U.S. has set a dangerous precedent about government control over American-made software releases.
Why it matters: The Trump administration just demonstrated it can unilaterally force any U.S. tech company to pull products offline without court approval — invoking an export control law designed for weapons to target AI models over behavior cybersecurity experts call legitimate. The precedent extends well beyond Anthropic to any software company that crosses the White House.
