Anthropic’s most powerful model, Mythos 5, has been approved to resume limited access after being restricted by the U.S. government for two weeks. On June 26, U.S. Secretary of Commerce Howard Lutnick confirmed that the model will be made available to approximately 100 vetted U.S. companies and federal agencies, but will not be publicly released in full.
This means the U.S. government has not withdrawn its security concerns regarding frontier AI, but has opted to allow access through a “trusted partner” framework. Rather than a complete ban, this resembles a controlled usage arrangement, signaling that Washington views high-capability AI as a technology requiring ongoing regulatory oversight.
Restricted within days of posting
Anthropic launched Mythos 5 on June 9, along with Fable 5, a version optimized for security purposes. A few days later, on June 12–13, the U.S. Department of Commerce imposed export controls on both models, forcing Anthropic to suspend customer access.
The official core reason cited is national security risks. The report notes that Mythos 5 can be used for drug design, vulnerability discovery, and biological defense screening—capabilities that have both civilian and sensitive applications, promptly triggering regulatory intervention.
Open only to approximately 100 institutions
According to Bloomberg, citing a letter from Lutnick to Anthropic executives, the U.S. government believes the company has taken improved measures to address related risks and has therefore permitted access to the model for "specific trusted partners."
Currently accessible entities primarily consist of approximately 100 vetted U.S. companies and federal agencies. Fable 5 remains publicly available, while Mythos 5 continues to be restricted.
From a timeline perspective, Anthropic regaining limited access within two weeks indicates that the company has reached more specific agreements with regulators on safety measures. The U.S. government also clearly recognizes that certain institutions have a real and urgent need for this model’s capabilities.
One million tokens in context is crucial.
One of the key capabilities of Mythos 5 that has drawn attention is its 1-million-token context window. According to the article, this enables the model to process large volumes of data in a single session—including extensive codebases, research documents, or complex rule texts—while maintaining coherent reasoning.
This capability goes beyond mere performance improvements. In high-risk scenarios, it enables the model to analyze complex systems more comprehensively, thereby enhancing its value in research, security, and infrastructure applications.
Has practical relevance to crypto security
For the crypto industry, the most direct connection lies in vulnerability detection capabilities. Smart contracts, cross-chain bridges, and DeFi protocols are fundamentally software systems, and related attacks often stem from logical flaws or hidden vulnerabilities in complex interactions.
If the model can analyze a broader range of code, it may have practical applications in auditing, protecting custodial systems, and securing on-chain financial infrastructure. The article suggests that this capability has the potential to address gaps in existing auditing tools and manual reviews.
However, the current access scope is still limited to institutions recognized by the U.S. government and does not include DeFi security firms or blockchain audit agencies. Whether this scope will expand in the future depends on how the U.S. government defines “trusted partners.”
Frontier AI regulation may set a precedent
The greater significance of this event is that the U.S. Department of Commerce has demonstrated a regulatory pathway: treating advanced AI models as "dual-use technologies" and incorporating them into the export control framework.
If this approach continues, the most powerful AI models in the future may remain in a tiered open-access state, with only a select few approved organizations gaining early access, followed by the general public and ordinary enterprises. Whether the outcome is a faster opening or prolonged restrictions, this will impact the pace of AI adoption in fintech, cybersecurity, and cryptocurrency infrastructure.