Trump administration restricts access to Anthropic’s Mythos 5 for foreign entities

Trump administration restricts access to Anthropic’s Mythos 5 for foreign entities

Export controls on Anthropic's latest AI models triggered a two-week standoff before limited access was restored for select US organizations

Anthropic’s most powerful AI model barely lasted three days before the US government pulled the plug. The Trump administration directed the company to suspend access to Mythos 5 and its companion model Claude Fable 5 for all foreign nationals on June 12, citing national security concerns. That was just days after Mythos 5 first went live on June 9-10 as part of a cybersecurity-focused initiative called Project Glasswing.

Rather than try to selectively restrict access, Anthropic took the nuclear option: it disabled both models entirely on June 12 to comply with the new export controls. The move effectively shut down access for everyone, not just the foreign nationals targeted by the Commerce Department’s directive.

Advertisement

A two-week standoff with high stakes

What followed was roughly two weeks of intense negotiations between Anthropic executives and Commerce Secretary Howard Lutnick. The export controls didn’t just apply to overseas users. They required licenses for foreign nationals working inside the US to access the models. In English: if your cybersecurity team includes engineers on work visas, they couldn’t touch Mythos 5 without government approval.

High-level correspondence between Lutnick and Anthropic leadership proved instrumental in breaking the impasse. Anthropic introduced additional safeguards, the specifics of which haven’t been fully detailed, and the Commerce Department agreed to lift the restrictions around June 26-30.

Limited access, limited audience

Access wasn’t restored to the general public or even to all paying customers. Instead, the government approved a narrow set of users: select US companies, federal agencies, and partners participating in Project Glasswing.

Mythos 5 is priced at $10 per million input tokens and $50 per million output tokens. The model’s primary strengths lie in cybersecurity testing and scientific hypothesis generation. Project Glasswing, the initiative through which Mythos 5 is primarily accessed, focuses on cybersecurity applications.

The resolution also established that foreign national employees at approved companies participating in Project Glasswing could regain access, addressing one of the core friction points that had made the models unusable for many organizations during the standoff.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.

Trump administration restricts access to Anthropic’s Mythos 5 for foreign entities

Trump administration restricts access to Anthropic’s Mythos 5 for foreign entities

Export controls on Anthropic's latest AI models triggered a two-week standoff before limited access was restored for select US organizations

Anthropic’s most powerful AI model barely lasted three days before the US government pulled the plug. The Trump administration directed the company to suspend access to Mythos 5 and its companion model Claude Fable 5 for all foreign nationals on June 12, citing national security concerns. That was just days after Mythos 5 first went live on June 9-10 as part of a cybersecurity-focused initiative called Project Glasswing.

Rather than try to selectively restrict access, Anthropic took the nuclear option: it disabled both models entirely on June 12 to comply with the new export controls. The move effectively shut down access for everyone, not just the foreign nationals targeted by the Commerce Department’s directive.

Advertisement

A two-week standoff with high stakes

What followed was roughly two weeks of intense negotiations between Anthropic executives and Commerce Secretary Howard Lutnick. The export controls didn’t just apply to overseas users. They required licenses for foreign nationals working inside the US to access the models. In English: if your cybersecurity team includes engineers on work visas, they couldn’t touch Mythos 5 without government approval.

High-level correspondence between Lutnick and Anthropic leadership proved instrumental in breaking the impasse. Anthropic introduced additional safeguards, the specifics of which haven’t been fully detailed, and the Commerce Department agreed to lift the restrictions around June 26-30.

Limited access, limited audience

Access wasn’t restored to the general public or even to all paying customers. Instead, the government approved a narrow set of users: select US companies, federal agencies, and partners participating in Project Glasswing.

Mythos 5 is priced at $10 per million input tokens and $50 per million output tokens. The model’s primary strengths lie in cybersecurity testing and scientific hypothesis generation. Project Glasswing, the initiative through which Mythos 5 is primarily accessed, focuses on cybersecurity applications.

The resolution also established that foreign national employees at approved companies participating in Project Glasswing could regain access, addressing one of the core friction points that had made the models unusable for many organizations during the standoff.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.