U.S. Commerce Secretary Howard Lutnick wrote to Anthropic chief compute officer Tom Brown on June 26 authorizing the company to restore access to Claude Mythos 5 for a defined set of around 100 organizations — a mix of federal agencies and private companies. Lutnick's letter states that 'appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,' and the approval is framed around defensive cyber and national-security use cases. Fable 5, the consumer-facing tier suspended at the same time, was not included in the order.
The action partially unwinds the June 12 export-control freeze under which Lutnick had directed Anthropic to cut off non-U.S. access to both Mythos 5 and Fable 5, citing distillation risk and adversary capability transfer. What's new is the mechanism: rather than a blanket reopening, the Department of Commerce is operating an allow-list model — trusted-partner approval, customer by customer, with the company doing the gating in concert with the government. The phrase 'safeguards are in place' is doing real work; it implies a verification regime around who can see the weights, the API surface, and the activity logs.
This is the first time a frontier-model release has been re-opened on a named-list basis, and it formalizes a posture that had been emerging across 2026: Washington wants to treat frontier inference like dual-use semiconductors. The June 2 White House executive order on advanced AI innovation and security set the framework — pre-release government access for cybersecurity and national-security review. Lutnick's letter is the first concrete application. Expect the same pattern with OpenAI's GPT-5.6 limited preview and any future Anthropic Fable tier.
A takeaway for learners: 'who can use this model' is now policy infrastructure, not just product strategy. If you are studying AI policy, watch for the standard operating procedure that grows around this letter — vetting criteria, audit logs, revocation clauses, and the diplomatic posture toward allies asking why their researchers were left off the list. If you are studying AI deployment, build for an environment where access lists, not just rate limits, govern frontier model availability.