Lifting Export Restrictions on Claude Mythos 5: Access for 100 US Organizations and a New Phase of the AI Race
On Friday, the U.S. Department of Commerce officially lifted the export ban on Anthropic's powerful artificial intelligence model, Claude Mythos 5. This decision grants access to the cutting-edge technology for over one hundred American organizations, including major corporations and government entities listed in the special Appendix A.
This move concludes a two-week standoff between the administration and the developer. As Commerce Secretary Howard Lutnick noted in a letter to Anthropic's technical director Tom Brown, existing safeguards are deemed sufficient to transfer the model to verified partners without requiring an additional license.
Path to Unblocking: From Project Glasswing to Global Accessibility
Until now, Mythos 5 was used exclusively within the Glasswing program — a vulnerability discovery initiative involving about 150 organizations from more than 15 countries. Notably, this model previously demonstrated impressive results, identifying errors in classified systems within just a few hours during government testing.
Export restrictions were imposed after one of Anthropic's largest investors, Amazon, expressed concerns about model safety. Company researchers warned that another powerful model, Fable 5, could potentially be hacked for dangerous purposes. Now, after the lifting of the blockade, the fate of Fable 5 remains in question.
Fable 5 Awaits: A New AI Regulatory Regime Takes Shape
Unlike Mythos, access to Fable 5 was previously open to all subscribers, and for a time it remained the most powerful public AI tool. However, its broad release is now postponed. According to sources close to the negotiations, the release process has moved forward, but specific timelines have not yet been announced.
The situation is gradually transforming into a new system of restrictions. On June 2, an executive order was signed, opening a voluntary channel for federal review of advanced AI models. Developers can submit their solutions for cybersecurity checks 30 days before launch. Simultaneously, Washington is tightening exports of AI chips to China, and now similar requirements are being extended to access the models themselves.
Significantly, OpenAI has followed the same path, restricting access to the most powerful version of GPT-5.6, named Sol, to about 20 partners who received government approval. Weaker versions, Terra and Luna, have become available to the general public.
Cryptalist Analysis: We are witnessing the formation of a fundamentally new control architecture in the AI field — not only over "hardware" but also over algorithms. For the crypto industry, where the security of smart contracts and DeFi protocols is critical, access to models like Mythos 5 could be a game-changer. But the key question is: will this new system of restrictions lead to fragmentation of the global cybersecurity market, where European and Asian companies become dependent on Washington's decisions?