The U.S. Commerce Department delivered an enforcement letter to Anthropic on Friday afternoon, invoking an obscure export‑control provision that effectively barred non‑American individuals—including the firm’s own engineers—from accessing its two flagship large‑language models, Fable 5 and Mythos 5. The department cited an unspecified national‑security concern but did not disclose further details, and the letter itself has not been released to the public.

In response, Anthropic immediately disabled both models for every customer, ensuring compliance with the directive. The company said it believes the action stems from a reported bypass of the models’ safety guardrails, though the agency’s notice offered no concrete explanation. By pulling the models offline without seeking court approval, the Commerce Department demonstrated a swift, unilateral capacity to intervene in the operations of a U.S. AI firm.

Security researcher Katie Moussouris, founder of Luta Security, described the episode as “hasty, heavy‑handed, and misguided.” Moussouris received a private copy of a paper authored by Amazon security researchers that detailed a guard‑rail bypass in Fable 5. The paper showed that the model could be coaxed into revealing code‑review advice when prompted to “fix” code, a subtle shift from a benign “review” request. Moussouris argued that such behavior should never have triggered an export‑control action, noting that the difference in phrasing does not meaningfully alter the model’s output.

Following the shutdown, dozens of cybersecurity experts and researchers signed statements urging the administration to revoke the order. They warned that removing advanced AI capabilities from U.S. defenders could be “dangerous” for national security, flipping the intended protection on its head. The episode revives memories of earlier export‑control regimes that, in the 2010s, cast a wide net over cybersecurity tools, unintentionally stifling legitimate research.

Justin Hendrix, editor of Tech Policy Press, warned that the move could erode confidence abroad in the reliability of American AI for critical applications. He suggested the government’s action sends a signal that U.S. AI companies cannot operate without the risk of sudden shutdowns driven by political or personal factors. The lack of transparency surrounding the Commerce Department’s rationale fuels speculation about whether the decision was a misreading of the research paper, a reaction to comments from industry leaders, or a broader attempt to pressure Anthropic, with whom the administration reportedly has a strained relationship.

Anthropic’s abrupt compliance highlights a new frontier in the intersection of AI development and export regulation. While the company has not disclosed how long the shutdown will last, it affirmed its commitment to cooperate fully with the government. The incident underscores the growing influence of U.S. policy on the global AI landscape, raising questions about how future innovations will navigate an increasingly complex regulatory environment.

Este artigo foi escrito com a assistência de IA.
News Factory APP - notícias agênticas para impulsionar seu SEO e AEO.