US removes export controls on Anthropic's Fable and Mythos AI models after three-week freeze.

The US Commerce Department has lifted export controls on Anthropic's advanced Fable and Mythos AI models, ending restrictions imposed less than three weeks earlier over national security concerns. Washington has intensified oversight of model releases to counter potential military applications of advanced AI in China, Russia, and other countries of concern.

On June 12, the Commerce Department ordered Anthropic to immediately restrict access to Mythos 5 and Fable 5 to US nationals only, citing national security risks. With no way to verify nationality in real time, Anthropic disabled both models for all users. Last week, the government permitted limited release of Mythos 5 to "trusted" US organizations. Mythos 5, designed to detect cybersecurity vulnerabilities as part of the broader Glasswing project, had previously been available to a wider group of companies.

Following implementation of new safeguards, all export controls have now been lifted. Anthropic said it is working with the US government to expand Mythos 5 access to a broader set of domestic and international Glasswing partners. Fable 5, intended for general public use and featuring stronger safeguards, became available from Wednesday.

The company is deepening collaboration with the US government and has granted designated partners expanded early access to both models. Working with Amazon, Microsoft, Google, and other Glasswing partners, Anthropic is developing shared standards to assess and remediate AI jailbreaks—techniques that bypass safeguards—and a system to rank their severity.

A letter from Commerce Secretary Howard Lutnick, seen by Reuters, confirmed Anthropic's agreement to work with the US government on protocols for these models and future releases, and to report any malicious activity. Lutnick cautioned that the department "reserves the right to reevaluate the decisions made in this letter and the necessity of reimposing a license requirement, should circumstances change or should Anthropic fail to adhere to its commitments."

The June 12 restrictions followed Amazon researchers' disclosure of a method to bypass Fable 5's safeguards, enabling the model to identify software vulnerabilities and generate exploitative code. Anthropic has implemented a new safeguard that blocks such behavior, redirecting blocked requests to its Opus 4.8 model instead. The company acknowledged this may frustrate users but represents a necessary tradeoff to preserve the model's broader capabilities.

Anthropic cautioned that making any AI model fully robust against jailbreaks is "probably impossible," and warned of potential universal jailbreaks capable of unlocking an entire class of harmful behaviors. "There will be many minor jailbreaks, some narrow harmful ones, and although no universal jailbreaks for Fable 5 have been discovered at the time of writing, expert safety researchers continue to red-team it," the company stated.

Isaac Harris, executive director of the Frontier Security Institute, noted that a formal standards process for US models now appears to exist, but raised concerns about asymmetric regulation: "There's still a question mark as to how equivalently dangerous capabilities coming from China with less guardrails will be handled by the administration in the US market."

Broader US scrutiny of AI models began this month when President Donald Trump signed an executive order establishing a voluntary framework requiring "covered frontier models" to be offered to the US government for up to 30 days before release to trusted partners. OpenAI has faced similar restrictions and delayed the full public launch of GPT-5.6 at the government's request, limiting access to a small group of vetted partners.

Anthropic's relationship with the US government has been contentious. The Pentagon designated the company a "supply-chain risk" earlier this year, barring contractors from using Anthropic's AI for military work, after the company declined to allow its models for mass domestic surveillance or fully autonomous weapons systems. Both OpenAI and Anthropic have confidentially filed for US initial public offerings.