Anthropic Export Controls: US Lifts Restrictions on Fable and Mythos AI Models

WASHINGTON: The United States has lifted export controls on Anthropic’s Fable and Mythos artificial intelligence models, less than three weeks after the company was ordered to suspend access to its most advanced systems over national security concerns.

Anthropic confirmed on Tuesday that the US Commerce Department had withdrawn restrictions and said access to the affected models would begin returning from the following day.

The company had disabled its Mythos 5 and Fable 5 models after receiving an export-control order on June 12. Earlier, US authorities allowed limited access to Mythos 5 only for selected trusted organisations inside the United States.

Kylian Mbappe: France Cruise Past Sweden to Reach World Cup Last 16

The latest decision marks a shift in Washington’s approach as officials continue increasing oversight of advanced AI systems amid concerns that powerful models could be exploited for military intelligence or cyber operations by foreign actors.

According to a letter from US Commerce Secretary Howard Lutnick, Anthropic agreed to strengthen efforts to identify and address security risks linked to its AI systems.

The company also committed to working closely with US authorities on security protocols covering Mythos, Fable and future AI releases, while notifying the government of any malicious activity involving its models.

However, US officials warned that the decision could be reviewed again.

The Commerce Department stated that it reserves the right to reimpose licensing requirements if circumstances change or if Anthropic fails to meet agreed commitments.

Although both models are built on the same underlying technology, Fable is designed for broad public availability, while Mythos includes fewer restrictions for specialised use cases.

A source familiar with the matter said Anthropic introduced an additional safeguard aimed at blocking techniques that could bypass protections in Fable 5.

The source added that vulnerabilities previously identified by regulators had already been addressed before the restrictions were lifted.

Experts have warned that highly capable AI models could significantly increase the sophistication of cyberattacks, particularly against sectors such as banking and critical infrastructure that depend on complex and ageing systems.

The easing of controls follows broader government efforts to shape AI deployment.

Last week, the US government also requested a limited release approach for OpenAI’s GPT-5.6, restricting early access to a smaller group of vetted partners.

Those moves prompted criticism from industry observers who questioned how authorities determine which organisations qualify as trusted participants.

Government oversight intensified after President Donald Trump signed an executive order creating a voluntary framework for developers of advanced AI models to provide early access to US authorities before wider deployment.

Security analysts said the latest decision suggests the United States is moving towards establishing clearer standards for advanced domestic AI systems.

At the same time, questions remain over how regulators will respond to comparable AI capabilities developed outside the United States.

Anthropic’s relationship with the US government has faced challenges this year. The Pentagon previously classified the company as a supply-chain risk after disagreements over the use of AI for mass surveillance and fully autonomous weapons systems.

Anthropic and OpenAI have both confidentially filed for initial public offerings in the United States.