The Trump administration continues to restrict Anthropic's strongest model, Mythos, making it inaccessible to overseas institutions.

The U.S. government has lifted export controls on Anthropic PBC's latest AI model while still restricting its more powerful Mythos 5 model for exclusive use by a small number of U.S. institutions, leaving global users—including foreign governments, businesses, and financial institutions—in a state of uncertainty.

According to Bloomberg, Anthropic stated that global users can begin using the Fable 5 model starting Wednesday, which is similar to Mythos and designed for broader release. However, negotiations over expanding domestic and international access to Mythos 5 through the "Glasswing Project" are still ongoing, with the U.S. government deciding which institutions will regain access.

An Anthropic spokesperson said the company is restoring access for original U.S. partners under the Glasswing Project, but there is no timeline for when international partners will regain access.

The lack of transparency from Washington regarding which institutions can use these models highlights the extent to which the digital security of many U.S. allies depends on U.S. policy. At the same time, it has made Anthropic a "test subject" for U.S. regulation of frontier AI models.

Export controls lifted, but Mythos access still restricted

Mythos is extremely capable at uncovering cybersecurity vulnerabilities. Anthropic previously restricted its access to a curated group of institutions to identify and patch vulnerabilities before malicious actors could exploit them. Anthropic stated that Fable 5 has comparable capabilities, but its built-in safety protections prevent the model from being used to expose cybersecurity vulnerabilities.

Anthropic has never publicly released Mythos. In April, the company opened access to an initial group of 50 institutions through the Glasswing Project, which expanded to 150 institutions across 15 countries by June.

This progress was interrupted by export control measures issued by the U.S. Department of Commerce on June 12, which required Anthropic to obtain U.S. government permission before allowing any foreign nationals (regardless of location) to access Fable or Mythos.

According to Anthropic, the control order was issued after reports surfaced that Amazon security researchers had found a way to bypass Fable 5's safety protections, allowing the model to be used for uncovering cybersecurity vulnerabilities.

The U.S. partially relaxed restrictions on Mythos last week and officially lifted the export controls on Tuesday. According to sources familiar with the matter, this shift came after Anthropic added a classifier (a safety filtering mechanism) to Fable 5.

U.S. approval conditions opaque, varied treatment for institutions across countries

In the previous open phase, access for institutions across different countries was uneven. The UK AI Security Institute was one of the few international organizations granted early access to test Mythos.

The European Union's cybersecurity agency ENISA has also been scheduled to gain access after weeks of negotiation. Japanese banking institutions were also included in the Mythos testing list, but European banking institutions did not receive corresponding access.

This uneven access landscape further reflects the dominant role of U.S. policy in determining who can use this cutting-edge tool, leaving overseas users, including financial institutions, continually subject to Washington's approval pace for cybersecurity protection capabilities.

Risk Warning and Disclaimer

        Markets involve risks. Invest with caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are appropriate for their specific circumstances. Investment based on this article is at your own risk.