The US government has partially reversed its ban on Anthropic’s most powerful AI models, allowing the company to restore access to its Mythos 5 system for more than 100 trusted American organisations. The move comes two weeks after the Commerce Department used national security export controls to disable the models entirely — a directive so broad that Anthropic had no choice but to shut them off for all users globally.
“Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organisations that operate and defend critical infrastructure,” Anthropic said in a statement on Friday. “We’re restoring access for these organisations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.”
Access will extend to many Fortune 500 companies, according to a source familiar with the directive. Many of the approved organisations are part of Anthropic’s Project Glasswing, which includes approximately 100 well-known technology companies and institutions. Under the new arrangement, an export licence will no longer be required for Mythos 5 to be used by trusted companies and their employees who are not US citizens — a significant concession from the initial order, which had also barred Anthropic’s own non-citizen employees from the models.
A letter from Commerce Secretary Howard Lutnick to Anthropic cited “significant progress” in work done between the company and the government to address “risks associated with the Covered Models,” though it did not specify what safeguards had been adopted. The government is also moving toward allowing Anthropic to release Fable 5 — the public-facing version built on the same underlying model as Mythos — though no timeline has been set.
ALSO READ: SpaceX Is Quietly Launching Starfall — a Secret Reentry Capsule That Could Own Orbital Manufacturing
The Concern and the Criticism
The original export control order stemmed from government concerns that Mythos’s advanced cybersecurity capabilities could be misused by military intelligence operators in China, Russia, or other countries of concern. The government reportedly learned of a technique to “jailbreak” Fable 5’s safeguards in a way that would unlock Mythos-level cybersecurity capabilities — a vulnerability Anthropic has said it believes is narrow and not meaningfully different from exploits that could be applied to other publicly available frontier models, including OpenAI’s GPT-5.5.
OpenAI separately confirmed Friday that it was delaying the full public launch of GPT-5.6 at the US government’s request, limiting access to a small group of vetted partners whose details were shared with authorities — a sign that the government’s new oversight posture extends beyond Anthropic alone.
The selective nature of the approval process has drawn sharp criticism from across the political spectrum. “No one knows how these companies are picked and why everyone else is excluded,” said John Coleman, legislative counsel for the Foundation for Individual Rights and Expression. “This is putting too much power in the hands of the government. There’s little transparency and it raises questions about the rule of law.”
OpenAI CEO Sam Altman echoed those concerns directly. Extensive safety testing “is not a bad idea,” he wrote on social media. “I just don’t like the idea of the government picking the customers.”
Analysts have raised a broader strategic concern. “The longer there isn’t a system in place that will allow US companies to widely release new models, the more likely it is that China will be able to catch up,” said Kate Koren, an analyst at the Center for Strategic and International Studies and former Commerce Department official.
Anthropic’s relationship with the US government has been particularly fraught. The company was placed on a Pentagon national security blacklist after refusing to allow its AI to be used for autonomous weapons systems or domestic mass surveillance — a position that put it in direct conflict with the Department of Defense’s preferred contract terms. The current restrictions follow an executive order Trump signed this month establishing a voluntary framework for AI developers to offer frontier models to the government for up to 30 days before broader release.
Stay informed. Subscribe to the JournalTodays Newsletter for the latest AI policy, technology news, and business coverage delivered straight to your inbox.
