The U.S. lifts the block on Anthropic's Mythos 5 model and releases it to more than 100 selected institutions

The article, published on June 26, 2026 by Reed Albergotti and Ben Smith in Semafor, reveals that the United States government formally lifted the block on Claude Mythos 5, Anthropic's most powerful artificial intelligence model, allowing access for more than 100 selected American institutions, including major companies and government agencies. The decision came through an official letter from Secretary of Commerce Howard Lutnick addressed to Tom Brown, Anthropic's chief compute officer, on Friday afternoon.

The unblocking marks a significant easing of the confrontation that had erupted just two weeks earlier, when the Trump administration imposed export controls on Mythos 5 and its sibling model, Fable 5. That measure effectively shut down both models after Amazon and other companies warned about the possibility that they could be 'jailbroken' for malicious purposes. The speed with which this agreement was reached—just fourteen days of intense, daily negotiations—is presented by Commerce Department spokesperson Benno Kass as a demonstration of the government's agility in protecting national security without sacrificing American technological leadership.

In his letter, Lutnick states that he has determined that 'appropriate safeguards exist to allow certain trusted partners to access the Claude Mythos 5 model,' citing 'significant progress' in the conversations between the administration and the company. The letter specifies that a license will no longer be required to export, re-export, or transfer within the country the model to the entities identified in Annex A, including their foreign employees and Anthropic's own foreign employees. In exchange, Anthropic has committed to collaborating with the U.S. government in designing protocols, standards, and release processes for its models.

Fable 5, presented in the article as a less powerful version of Mythos—and which briefly became the most powerful model available to consumers in general—is not mentioned in Lutnick's letter. However, sources close to the negotiations cited by Semafor indicate that conversations are also moving toward its release, although the timeline remains uncertain. This omission is significant: it implies that mass consumer access is not yet guaranteed and that the government is maintaining, for now, tighter control over the most capable and publicly widespread models.

On the same day as the announcement, Anthropic's direct competitor, OpenAI, released its latest model, GPT-5.6, to a small list of government-approved partners. This timing suggests that a new regulatory framework is taking shape in which the U.S. administration plays the role of arbiter and gatekeeper in the release of frontier artificial intelligence models—something that until very recently was unthinkable in the American tech sector, historically reluctant to embrace government intervention.

Semafor notes that this episode marks the beginning of a new regulatory regime in which the United States government takes control over the release of frontier AI models. Although the leaders of the major labs have shown resistance to the prospect of losing time in an extremely competitive global technology race, the Commerce Department argues that acting this quickly proves that security and leadership are not incompatible goals.

One of the triggers of the original block, as Semafor itself previously reported exclusively, was the U.S. government's concern that Mythos had been made available to partners too closely tied to China. Reports specifically pointed to a South Korean telecommunications provider as a concrete case that raised alarm in Washington. This geopolitical element adds an additional layer of complexity to the episode: it is not only about technical concerns over the model's security, but about a broader battle for control over who accesses the world's most advanced AI technologies and under what conditions.

The article also focuses on the collateral effects of this new regulatory dynamic on international allies and partners. European governments and other nations allied with the United States have expressed frustration at their new dependence on decisions made in Washington in order to access AI tools they already considered part of their habitual technology ecosystem. Both non-U.S. users and companies and governments outside the Annex A list remain uncertain about when and under what conditions they will be able to access Mythos and Fable.

From a broader perspective, this episode represents a milestone in the evolution of artificial intelligence governance. For the first time, the government of a democratic country has actively and explicitly exercised its authority to block and then re-authorize the release of a private frontier AI model, based on national security and export-control criteria. The legal architecture used—export controls—is the same one that has been employed to limit the sale of advanced chips to China, indicating that the administration is extending its technology regulatory arsenal into the realm of software and AI models.

The fact that the letter speaks of 'trusted partners' and of an 'Annex A' with specific entities suggests that a whitelist system for access to advanced AI is being born, with criteria that are not yet fully transparent to the public or to the international community. This raises fundamental questions about the future of open access to AI models, the competitiveness of companies and governments not included on those lists, and the role that geopolitical alliances will play in the artificial intelligence ecosystem in the coming years.

In short, the Semafor article documents a turning point in the relationship between the major AI labs, the U.S. government, and the rest of the world: the birth of a regime of government oversight over frontier AI that, although built 'on the fly' according to the text itself, sets a precedent with far-reaching consequences for the entire industry.