After weeks of intensive negotiations, the White House lifted a key restriction on Anthropic’s flagship AI system, Claude Mythos 5. In a letter obtained by WIRED, Commerce Secretary Howard Lutnick told Anthropic co‑founder and chief compute officer Tom Brown that the company could grant access to the model for a select group of more than 100 U.S. corporations and government agencies. The move marks the first major easing of a June‑12 export‑control directive that had forced Anthropic to shut down all external access to its most powerful models.

Lutnick’s letter emphasized that the approved partners have demonstrated “appropriate safeguards” to mitigate the risks the administration cited when it first acted. Those risks included the possibility that foreign nationals—particularly those with ties to China—could use the model to develop cyber‑offensive capabilities, as well as concerns raised by the National Security Agency and Amazon that the consumer‑facing version, Claude Fable 5, could be jail‑broken.

Anthropic’s response was swift. Spokesperson Eduardo Maia Silva confirmed the company would begin provisioning the cleared providers and restore their access to Mythos 5 “as quickly as possible.” He added that Anthropic remains in talks with the White House about reinstating broader access to Fable 5, a discussion expected to continue over the weekend.

The original directive, issued under the Trump administration, required Anthropic to block foreign nationals—whether working for U.S. firms or residing in the United States—from using either Mythos 5 or Fable 5. In compliance, the company disabled the models entirely, prompting a legal confrontation earlier in the year when Anthropic sued over a supply‑chain risk designation. The latest letter relaxes those restrictions for the approved U.S. entities, allowing them to let foreign‑national employees engage with Mythos 5, and granting Anthropic the same latitude for its own staff.

Background concerns stemmed from a June incident in which Anthropic reportedly granted access to a South Korean telecom firm believed to have connections to China. That episode, combined with internal warnings from the NSA and external pressure from industry players, convinced officials that a hard stop was necessary until clearer safeguards could be demonstrated.

During the negotiations, senior members of Anthropic’s cybersecurity and AI‑safety teams traveled to Washington, D.C., meeting with officials from the Department of Commerce and other agencies. Alongside Brown, public‑policy chief Sarah Heck helped steer the dialogue, emphasizing the company’s commitment to responsible AI deployment.

While the partial reinstatement of Mythos 5 is a tangible win for Anthropic, it also underscores the broader tension shaping U.S. AI policy. The administration’s willingness to grant limited access suggests a calibrated approach—allowing critical infrastructure and defense partners to benefit from cutting‑edge models while keeping a tighter leash on broader consumer releases.

OpenAI, another front‑runner in the AI space, recently announced a delay to its upcoming GPT 5.6 models after a separate request from the Trump administration, highlighting that Anthropic’s episode is part of a larger regulatory push. Dean Ball, head of OpenAI’s strategic futures team and former White House AI adviser, warned that developers now need an “explicit green light from the government” before moving forward with frontier models.

Anthropic’s investors spent the weekend assessing the impact of the latest developments on the company’s trajectory. Though the partial lift does not yet restore full market access, the company views it as a step toward a lasting policy framework that could eventually allow both Mythos 5 and Fable 5 to operate at scale under government‑approved safeguards.

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