US clears Anthropic's Mythos 5 for 100 trusted institutions

For two weeks, access to Anthropic's most powerful model was effectively under federal lock and key. On June 26, Semafor reported that Commerce Secretary Howard Lutnick had notified Anthropic that it could again make Claude Mythos 5 available to a group of roughly 100 US companies and federal agencies, writing that he had "determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model."

The backstory matters for understanding what kind of precedent this sets. On June 12, the government had invoked export control authorities to require Anthropic to suspend access to both Mythos 5 and Fable 5, its two most capable models, citing a reported jailbreak and national security concerns. Anthropic pushed back on the severity of the finding, writing that "we suspect that perfect jailbreak resistance is not currently possible for any model provider," and characterizing the vulnerabilities found as previously known and minor.

The resolution established a new structure rather than simply restoring the status quo. The Lutnick letter eliminated export licensing requirements for approved entities listed in an annex and their foreign national employees, and Anthropic committed to work with the US government on protocols, standards, and model releases going forward. Before the ban, Mythos 5 had been accessible to a subset of trusted organizations in Anthropic's Project Glasswing, including infrastructure providers like Cisco and banks like JPMorgan Chase, and that cohort appears to be the rough template for the restored access list.

The honest caveat is that the reporting leaves key details unresolved. Lutnick's letter does not mention any change in restrictions for Fable 5, leaving that model in limbo, and the reporting does not explain what "appropriate safeguards" Lutnick found satisfactory or who exactly makes up the roughly 100 approved institutions.

The practical shape of this is a government-controlled access tier for frontier AI, with Washington as gatekeeper. For enterprises and agencies that make the list, the two-week disruption is over. For those outside the annex, this is a signal that even commercially available frontier models can be pulled on short notice, and that a company's relationship with the government may increasingly determine access.