Lifting Export Restrictions on Claude Mythos 5: A New Turn in the AI Safety Race
The U.S. administration has officially lifted the ban on exporting the Claude Mythos 5 model from Anthropic. This decision, made on Friday, opens access to the advanced AI system for over 100 organizations, including major corporations and government agencies. This concludes a two-week standoff between the Trump team and the developers.
Breakthrough for the select few: Mythos 5 returns
Commerce Secretary Howard Lutnick officially notified Anthropic CTO Tom Brown of the removal of licensing requirements for transferring the model to organizations on the "Appendix A" list. In his letter, he emphasized that current safeguards allow trusted partners to access Mythos 5 without additional bureaucratic hurdles.
Recall that both models—Mythos and Fable—were previously disabled after one of Anthropic's key investors, Amazon, expressed serious concerns about their safety. Company researchers warned that Fable 5 could be hacked for potentially dangerous purposes. Until then, Mythos was used exclusively within the Project Glasswing program—a vulnerability discovery initiative involving about 150 organizations from over 15 countries. Notably, within a few hours of testing, Mythos identified critical errors in classified government systems.
Fable 5 in limbo: New rules of the game
The fate of the public version of Fable 5 remains uncertain. Unlike Mythos, access to this model was previously open to all subscribers, and for a time it was considered the most powerful public AI tool on the market. However, its release process is now frozen, with no exact timeline provided.
The situation unfolds against the backdrop of a new regulatory framework. On June 2, an executive order was signed introducing a voluntary federal review channel for advanced AI models. Developers can submit their solutions for cybersecurity checks 30 days before launch. Over the past year, Washington has tightened exports of AI chips to China, and similar requirements now extend to access to the models themselves.
Notably, OpenAI has followed the same path: on Friday, the company restricted access to the most powerful version of GPT-5.6, named Sol, to about 20 state-approved partners. Weaker versions, Terra and Luna, became available to the general public.
Initially, the blocks were driven by concerns that Chinese entities might gain access to the technologies. South Korean operator SK Telecom came under scrutiny, being added to the Glasswing list in early June and then having its access restricted. SK Telecom categorically denied any ties to China.
Many leaders of major cybersecurity companies, including former Facebook security chief Alex Stamos, as well as representatives from Nvidia, Adobe, and Zoom, urged authorities to drop the restrictive measures. European countries and other regions found themselves heavily dependent on Washington's decisions regarding access to new developments.
My analysis: Lifting restrictions on Mythos 5 is not just a victory for Anthropic but a signal of a new AI regulatory model taking shape. The U.S. is moving from total bans to a "trusted partner" system, setting a precedent for the future market. The only question is whether Fable 5 will receive similar approval—and how quickly. If so, we will witness a new era in the AI arms race, where access to the most powerful models will be determined not by technological but by political criteria.