In a formal correspondence directed to Anthropic’s cofounder and chief compute officer, Tom Brown, US Commerce Secretary Howard Lutnick communicated that the AI firm is now permitted to grant access to 'Mythos' to specific trusted partners. This authorization follows a determination by the government that appropriate safety safeguards are now firmly in place.
Assessment of Safety Protocols
Lutnick emphasized in his communication that Anthropic has collaborated effectively with the US government to mitigate risks associated with the so-called Covered Models, leading to substantial advancements. Despite this progress, the administration stopped short of authorizing a wider rollout of the AI system. Furthermore, no clarification was provided regarding the status of Claude Fable 5, the consumer-facing version of Mythos that launched with additional security layers. The letter stipulated that all requirements outlined in the directive issued on June 12 remain in full effect.
Restoration of Mythos 5 Access
Eduardo Maia Silva, a spokesperson for Anthropic, stated that the company received notice from the US government confirming that Mythos 5, their most robust cybersecurity model, can be redeployed to a limited cohort of cyber defenders and infrastructure providers. Silva noted that the company is currently working to provision these approved providers and restore their access to Mythos 5 as swiftly as possible. Anthropic remains optimistic about this development and intends to continue collaborating with the government to expand access to Mythos 5 and eventually restore general availability for Fable 5.
Adjustments for Foreign National Personnel
This partial reinstatement arrives roughly two weeks after the White House issued an export control directive requiring Anthropic to restrict access to Claude Mythos and Fable 5 for foreign nationals, including those currently living and working within the United States. In the most recent letter, Lutnick clarified that organizations authorized to utilize Mythos are now permitted to grant access to their foreign national employees, and Anthropic is authorized to extend similar access to its own employees of foreign nationality.
Diplomatic Tensions and Context
The Trump administration's scrutiny of Anthropic intensified following reports that the company had provided model access to a South Korean telecommunications firm perceived to have links to China. Concerns raised by Amazon and the National Security Agency regarding the potential for Fable 5 to be compromised through jailbreaking prompted the administration to take decisive action. In recent weeks, Anthropic dispatched senior members of its cybersecurity and AI safety teams to Washington, DC, to engage with administration officials. Alongside Tom Brown, the company’s public policy chief, Sarah Heck, has spearheaded these discussions with the US Department of Commerce.
The Trajectory of US AI Policy
While the return of Mythos 5 represents a promising development for both Anthropic and the White House, the ongoing situation has sparked broader discourse regarding the direction of US AI policy and the extent to which the Trump administration will regulate future model launches. On Friday, OpenAI disclosed it was delaying the debut of its upcoming GPT 5.6 models following a request from the Trump administration. Dean Ball, head of the strategic futures team at OpenAI and a former White House AI adviser, noted in a June 16 blog post that the friction between Anthropic and the White House signals to frontier AI developers that they now require an explicit green light from the government to proceed. These regulatory battles have created financial and operational pressure for the startup. Earlier this year, Anthropic initiated litigation against the Trump administration regarding a supply chain risk designation received after attempting to establish boundaries for military contractors using its models. Following the export directive on the evening of June 12, stakeholders and investors were forced to quickly reassess the company’s corporate outlook.
