U.S. to lift export controls on Anthropic’s Mythos 5 and Fable 5 models

The U.S. Commerce Department informed Anthropic late Tuesday that it will lift export controls on the company’s Mythos 5 and Fable 5 models, allowing the company to restore broader access to the systems. Both Anthropic and the Commerce Department confirmed the decision the same evening. Anthropic stated it would begin restoring access the following day and thanked users for their patience.

The export-control order, originally imposed on national security grounds, had restricted access to the models and sparked confusion within the AI and cybersecurity communities. The move raised questions about what specific characteristics of the models were deemed uniquely risky. Earlier in the year, Anthropic released the first Mythos variant through Project Glasswing, a limited-access initiative focused on providing cyber-AI models to trusted organizations primarily for cyberdefense purposes.

Fable 5, released in early June, is the broader-access, safeguarded version of Anthropic’s Mythos-class model, while Mythos 5 is the restricted Project Glasswing version available to vetted groups with some cybersecurity safeguards removed. The National Security Agency (NSA) was among the organizations affected by the export control order and had been using the latest Mythos build.

The Trump administration had partially lifted the ban on Mythos 5 the previous Friday, allowing around 100 organizations to regain access, but kept restrictions on Fable 5 in place until this week’s decision. The reversal comes as Chinese open-source models begin to demonstrate cyber capabilities comparable to those of major U.S. AI labs.

Anthropic acknowledged a report by Amazon researchers, which claimed to have found a way to bypass Fable 5’s safeguards to identify exploitable software flaws. The company disputed the significance of the report, asserting that the bypass did not grant general users access to the more sensitive cyber capabilities reserved for Mythos. Anthropic said it trained a new safety filter that blocks the reported workaround more than 99% of the time.

Anthropic is collaborating with Amazon, Microsoft, Google, and other Glasswing partners on a common framework for assessing AI jailbreaks. This includes determining when a bypass is serious enough to require new safeguards or other actions from model developers.

The export-control standoff is unfolding amid an ongoing legal dispute between Anthropic and the Pentagon. The Pentagon designated Anthropic as a 'supply-chain risk,' a move Anthropic has argued was retaliation for its refusal to relax limits on certain military uses of its models. A federal judge granted Anthropic a preliminary injunction in March blocking parts of the designation and a subsequent order to end all government use of its products, though the government has appealed, and litigation continues.

The lack of clarity around the export directive has influenced how other AI firms approach model deployments. OpenAI recently announced it will initially limit access to three of its GPT-5.6 models after discussions with government officials, while testing the systems with select partners before broader release.