The U.S. government has partially lifted the export ban on AI firm Anthropic’s frontier models. The move allows a select group to regain access to the Claude Mythos 5 cybersecurity model, but the second frontier model, Fable 5, remains blocked — for now — with discussions ongoing.

The news was confirmed in a letter from Commerce Secretary Howard Lutnick to Anthropic in a letter dated June 26, according to a letter obtained by POLITICO .

The move comes as ChatGPT owner OpenAI, one of Anthropic’s largest competitors, confirmed it had restricted its cybersecurity AI model after the U.S. government raised concerns about its capabilities.

However, OpenAI said it is working towards wider availability. “At their request, we are starting with a limited preview for a small group of trusted partners whose participation has been shared with the government, before releasing more broadly,” OpenAI said in a blog post .

OpenAI explained how it is working towards broader availability of its frontier AI models. “We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them.”

Anthropic Versus The U.S. Government

There are some really interesting outcomes from the recent banning back of the latest AI model from Anthropic, says Ian Thronton-Trump, CISO at Inversion6. He thinks think part of it is political, “as the relationship between the U.S. government and Anthropic has not been a smooth one.”

Indeed, as POLITICO reports, there has been concern that Trump’s aides were unfairly singling out Anthropic after clashing with the firm previously.

There is, of course, a national security dimension to this, Thornton-Trump says. “As the U.S. is by far and away the largest but also the most vulnerable digital economy.”

At the same time, there are foreign policy considerations concerning China's rapid development. “It's clear that in the fast-paced world of AI model development an AI arms race is underway,” Thronton-Trump explains.

For example, Chinese researchers claim that Chinese AI startup Zhipu AI’s latest open-weight model, GLM-5.2, can match Anthropic’s flagship Mythos model in certain bug-finding tasks. “So it means the AI industry — already volatile — is now prone to more scrutiny."

“It’s too early to say if that is a good or bad thing, but it will be highly disruptive,” predicts Thornton-Trump. "Markets, venture capitalists and private equity investors want predictability and the AI market right now is anything but that.”

AI Models Abruptly Removed

The move to limit AI use by foreign nationals by restricting Anthropic’s Claude Mythos and Fable 5 took place in early June amid concerns over national security. The frontier AI models are able to find security vulnerabilities at scale, and this led to worries that attackers could also use them for this purpose.

Yet it meant the models were abruptly removed from European countries.

It is with this in mind that Austria has asked the EU to consider hosting Anthropic to preserve access to advanced AI models that have recently been restricted by the U.S.

Bill Connor, president and CEO of Jitterbit, a former adviser to U.S. presidents and U.K. prime ministers, thinks the government’s directive “highlights an important early learning phase for data sovereignty in the AI era.”

However, he points out the wider risk of policies such as this. “What global businesses need is not a policy that any one administration can live with. They need durable, jurisdiction-agnostic frameworks that do not shift when the political wind changes.”

When Anthropic first announced Mythos in April, it warned that the model was too capable for general release. As the model was limited by the U.S. administration, you could argue that it became a victim of its own hype. Yet in an era where AI development is moving so rapidly, guardrails do need to be in place. Striking the balance between safety and innovation is important and likely to continue to be an ongoing battle.