- US has withdrawn export controls on Anthropic's Fable 5 and Mythos 5 AI models.
- Anthropic will begin restoring access after agreeing to work with the US government on security measures.
- The episode signals a tougher regulatory approach for advanced AI models in the future.
The US government has lifted export restrictions on Anthropic's advanced AI models, Fable 5 and Mythos 5, ending weeks of uncertainty over whether the company's latest technology could be made available outside the country.
The decision allows the AI company to restore access to both models after they were suspended in mid-June following concerns that the technology could pose national security risks if it reached hostile governments or organisations. The move is another sign that AI regulation is beginning to extend beyond computer chips and into the software itself, with the US taking a more direct role in deciding how powerful AI systems are deployed.
Announcing the development on X, Anthropic said it had received formal notice from the US Department of Commerce that export controls had been withdrawn. The company said it would begin restoring access on July 2 and thanked users for their patience while the restrictions were in place.
The restrictions were first imposed on June 12, making Anthropic disable access to its newly launched Fable 5 and Mythos 5 models. A partial relaxation followed on June 28, when the US government allowed Mythos 5 to be used by selected trusted organisations inside the country. The latest decision removes the requirement for export licences altogether.
A compromise that could shape future AI releases
According to a letter from US Commerce Secretary Howard Lutnick, as quoted in a news report, Anthropic agreed to actively identify and address security risks linked to the models, work with the US government on safety protocols for future AI releases and report any malicious activity involving the technology. Based on those commitments, the government withdrew the export controls.
The decision comes amid growing concerns in Washington that increasingly capable AI models could be exploited by military or intelligence agencies in countries such as China and Russia. It is the first major case in which the US has used export control laws directly against AI models rather than semiconductor chips or computing hardware, potentially creating a new regulatory framework for frontier AI systems.
The episode also raises broader questions for the AI industry. OpenAI chief executive Sam Altman reportedly said he supports rigorous safety testing but does not favour governments deciding which customers should gain access to advanced AI models.
OpenAI has also faced similar scrutiny. The company recently delayed the wider rollout of GPT-5.6, limiting access to vetted partners following discussions with the US government. While the circumstances differ, the move suggests closer government oversight may increasingly become part of launching cutting-edge AI systems.
A difficult relationship with Washington
Founded in 2021 by siblings Dario Amodei and Daniela Amodei along with former OpenAI executives, Anthropic has had a complicated relationship with the Donald Trump administration.
Earlier this year, the US Department of Defense cancelled a contract worth £147 million ($200 million) after Anthropic declined to allow its AI models to be used for mass domestic surveillance or fully autonomous lethal weapons. The company maintained that current AI technology was not reliable enough for battlefield deployment without strict safeguards.
Following the disagreement, the Pentagon reportedly classified Anthropic as a supply chain risk and later signed a separate agreement with OpenAI for similar classified military capabilities.
Although Anthropic has regained access to international markets for its latest AI models, the episode suggests governments are becoming more willing to intervene in how advanced AI is released. For AI developers, meeting national security expectations could increasingly become as important as technological breakthroughs.
