A significant shift in US artificial intelligence regulation occurred on Friday. Anthropic's Claude Mythos 5 model received official export authorization for over 100 American organizations. This involves major corporations and government agencies that can now access this powerful AI system without needing an individual license.

This decision resulted from a two-week standoff between the Trump administration and Anthropic. Commerce Secretary Howard Lutnick, in a letter to the company's technical director Tom Brown, explained that current safeguards are deemed sufficient for transferring the model to verified partners from Appendix A. Essentially, this means the two-week export ban has been lifted, and Mythos 5 is back in operation.

What does this mean for the AI market and crypto security?

Recall that Mythos was previously used exclusively within Project Glasswing — a vulnerability discovery program involving approximately 150 organizations from over 15 countries. Within hours of testing, this model was already finding critical errors in classified government systems. Now that access is open to a wide range of American companies and agencies, we can expect significant acceleration in identifying and fixing vulnerabilities, including those in crypto infrastructure.

However, the situation with Fable 5, Anthropic's most powerful public model, remains uncertain. Unlike Mythos, Fable 5 was previously available to all subscribers but was then disabled due to concerns from Amazon, Anthropic's largest investor, about its safety. Researchers warned that Fable 5 could be hacked for potentially dangerous purposes. Now that restrictions have been lifted for Mythos, all attention is on Fable 5.

New AI regulatory regime

Concurrently, an executive order was signed on June 2, creating a voluntary channel for federal review of advanced AI models. Developers can submit their solutions for cybersecurity checks 30 days before launch. This is part of a broader trend: Washington is tightening not only the export of AI chips to China but also control over access to the models themselves.

OpenAI has followed the same path, restricting access to the most powerful version of GPT-5.6, called Sol, to approximately 20 state-approved partners. Weaker versions, Terra and Luna, have become available to the general public.

Notably, the initial trigger for the blocks was concern that the technology could gain access through South Korean operator SK Telecom, which was added to the Glasswing list in early June and then had its access restricted. SK Telecom denied any connection to China, but the suspicion remains.

Cybersecurity leaders, including former Facebook security chief Alex Stamos, Nvidia, Adobe, and Zoom, urged authorities to abandon restrictive measures. Now that Mythos 5 is unlocked, the question is whether Fable 5 will receive similar authorization.

My analysis: This precedent sets a new standard for the entire AI industry. We are witnessing the formation of a two-tier access system: "safe" models for the general public and "super-powerful" ones only for state-verified organizations. For the crypto industry, this is a dual signal: on one hand, access to the best tools for security auditing; on the other, increased control over technologies that could be used to hack smart contracts and DeFi protocols.