U.S. Authorizes Access to Anthropic's Mythos 5 After Two Weeks of National Security Block

U.S. Commerce Secretary Howard Lutnick signed a letter on Friday that partially lifts the most significant blockade ever imposed on a commercial artificial intelligence model by a government. "I have determined that appropriate safeguards are in place to allow specific trusted partners access to the Mythos 5 model," Lutnick wrote, according to Semafor. The letter was addressed to Anthropic co-founder Tom Brown, and it authorizes access for over 100 U.S. companies and federal agencies to the model, as reported by CNBC and Semafor.

The Mythos 5 has been offline for approximately two weeks following the Trump administration's export controls imposed on June 12. This action came after Amazon CEO Andy Jassy alerted Treasury Secretary Scott Bessent about jailbreak vulnerabilities in Anthropic's models. The directive to the company was to suspend access for "any foreign national, inside or outside the United States, including foreign employees of Anthropic," according to reports from CNBC and Semafor. The Fable 5, the less powerful version, does not appear in Lutnick's letter and remains restricted.

Outside the 100 Authorized, the Blockade Continues

Companies outside the United States that access Anthropic models via AWS Marketplace and Google Cloud continue to lack access to the Mythos 5. In Brazil, groups in the financial sector and IT consulting using AWS infrastructure for generative AI solutions face the same situation: access to Mythos 5 is unavailable. Anthropic distributed its previous models through partnerships with Amazon Web Services, which received a $4 billion investment from Amazon in 2023, and Google Cloud. Both platforms implement the access filter as defined by the U.S. government, and the public criteria for determining who qualifies as a "trusted partner" have yet to be disclosed.

Anthropic had not issued a public statement regarding the clearance by the time this report was published.

The Week When Two Frontier Models Came Under Federal Control

The Mythos 5 was not alone at the center of U.S. AI governance. OpenAI launched GPT-5.6 Sol, Terra, and Luna on June 26 in a preview restricted to approximately 20 partners vetted by the White House, according to a report from Axios. These two simultaneous moves, involving competing companies under different federal control regimes, indicate that the Trump administration treats the approval of frontier models as part of a regulatory layer exercised by export control authority, without a codified basis in AI law passed by Congress. The involvement of Amazon's CEO in the process leading to the blockade of Mythos 5 sets an additional precedent: hyperscalers informing the government about vulnerabilities in models where they are simultaneously investors and distributors.

For banks and consultancies building products on American frontier models, the signal of the week is clear: access can be suspended without prior notice and without a defined end date, by administrative decision without a basis in AI law passed by Congress.

The European Counterpoint and Transatlantic Dependence

In Europe, the European Parliament approved the Digital Omnibus on June 16, which postponed the AI Act requirements for high-risk systems in Annex III from August 2026 to December 2027. The regulatory easing in Europe coincides with the tightening of controls in the U.S. For banks like Deutsche Bank and HSBC, and for consultancies like Capgemini, which operate under both European regulation and dependence on American models, the two movements do not offset each other: one alleviates compliance bureaucracy; the other introduces the risk of access disruption to models without predictable legal basis.

In Japan, large banks and IT consultancies that have built solutions on platforms like AWS and Google Cloud are awaiting clarification on how U.S. export controls apply to existing contracts. The regulated Japanese sector lacks its own mechanism to respond to restrictions imposed by Washington without warning, a gap that the Mythos 5 episode has made apparent.

The question that the market still cannot answer: when the next frontier model from Anthropic arrives, will the list of "trusted partners" authorized by the U.S. government include or exclude companies that compete with the hyperscalers distributing these models?