The US Department of Commerce has partially reversed its export ban on Anthropic's most capable AI models, allowing Claude Mythos 5 to be accessed by a list of more than 100 vetted American companies and government agencies. The partial reinstatement, confirmed on June 26, covers non-US employees at those organizations as well — a notable carve-out given that the original ban explicitly included foreign nationals working inside the United States, including Anthropic's own non-citizen staff.

Fable 5, the more widely released version of the same underlying model, remains under the export control directive and unavailable to the public.

How the ban began

The Commerce Department issued the initial export control directive on June 12, citing national security authorities after government officials said they had identified a method to jailbreak Mythos 5 in ways that could enable significant harm. The directive suspended access for any foreign national — anywhere in the world, including inside the US — to both Fable 5 and Mythos 5.

Anthropic was unable to comply selectively. Verifying users' nationality in real time across multiple cloud platforms — AWS Bedrock, Google Cloud, Microsoft Foundry, Snowflake, and its direct Claude APIs — was not technically feasible on short notice. The company shut down access globally for all customers within hours, and posted the government's directive publicly on its website.

The partial reversal

Commerce Secretary Howard Lutnick's department determined that appropriate safeguards were in place to permit a subset of trusted partners to access Mythos 5. The list runs to more than 100 entities, including US government agencies and vetted private organizations, and extends to those organizations' foreign national employees — a meaningful shift from the June 12 directive that treated foreign nationality as an absolute bar regardless of employment context.

The restoration of Mythos 5 came the same day OpenAI previewed GPT-5.6 Sol to a similarly restricted set of government-approved partners. Together, the two decisions mark a rapid consolidation of a new framework for frontier model releases in the US: advanced models reach the public through a filter of government-sanctioned previews and approved-entity lists rather than open availability.

Fable 5 remains restricted

Fable 5 — the version Anthropic described as having additional safety protections and which it released publicly two days before the June 12 ban — has not been addressed in the partial reversal. Anthropic said in a statement that it continues to engage with the administration on restoring access for the broader user base. According to subsequent reporting, those discussions are ongoing and the company is cautiously optimistic, though no timeline has been set.

The episode has drawn attention to what voluntary and mandatory government oversight looks like in practice. OpenAI's arrangement was negotiated in advance and framed as a voluntary preview process. Anthropic's was a mandatory halt — imposed without advance notice and reversed only partially. Both companies have publicly stated that they view the current arrangements as temporary and not a model they would support as a permanent framework for model distribution.