Every international user of Anthropic’s two newest AI models just got locked out. The Trump administration declared Claude Fable 5 and Claude Mythos 5 security risks, imposing export controls that effectively cut off access for anyone outside the US or its approved partner network.
The timing is remarkable. Anthropic launched both models on June 9, 2026, positioning them as the company’s most capable AI systems to date. Days later, the government stepped in, requiring Anthropic to obtain permits for any deployment outside US borders or to foreign entities.
What happened and why it matters
Anthropic released two distinct versions of its latest technology, each serving a different purpose. Fable 5 is the public-facing model, built with enhanced safety guardrails designed to prevent harmful outputs. Mythos 5 is the restricted sibling, a more powerful cybersecurity-focused model made available only to vetted users through a program called Project Glasswing, which includes US government cyber defenders.
Mythos 5 ships without the safety guardrails that constrain Fable 5, making it far more capable for offensive and defensive cybersecurity operations. That plan unraveled when another company reportedly managed to jerry-rig Mythos to bypass its security features, raising immediate alarms in Washington.
Rather than targeting just Mythos 5, the export controls swept up Fable 5 as well, cutting off all international users from both models simultaneously. The only exceptions are designated Project Glasswing partners, a narrow group of approved entities working on national security applications.
The financial fallout for Anthropic
The pricing on both models had already doubled compared to previous generations. Anthropic set costs at $10 per million input tokens and $50 per million output tokens, twice what the earlier Opus 4.8 models charged. The export controls effectively limit Anthropic’s revenue potential to domestic clients and the narrow circle of Glasswing-approved partners.
Competitors like OpenAI, Google DeepMind, and emerging players in Europe and Asia now have an unexpected opening. International customers who might have chosen Anthropic’s models will need alternatives, and availability tends to win over superiority when organizations need to ship products.
The bigger picture for AI regulation
The dual-model approach Anthropic tried — one safe version for the public and one unrestricted version for trusted partners — was essentially a test of whether the industry could self-regulate access to its most powerful tools. The jailbreak of Mythos suggests that test failed, at least in the government’s eyes.
This is the US government treating frontier AI models the way it treats advanced semiconductor technology. Any company developing frontier models now has to consider the possibility that the government could impose export controls on their products at any time, based on demonstrated or even theoretical security vulnerabilities.
What this means for investors
Anthropic doubled its token costs from Opus 4.8 to the Fable 5 and Mythos 5 generation, and now has a much smaller market to sell into at those higher prices. If competitors don’t face the same restrictions, Anthropic could find itself charging premium rates to a shrinking customer base while rivals capture the international demand it can’t serve.
Venture investors have been pricing AI companies based on total addressable markets that assume global reach. If the US government is willing to unilaterally restrict that reach based on security concerns, the risk premium on AI investments just went up, and the discount rates applied to future cash flows should follow.