Trump administration export-control directive allows select release of Anthropic’s Mythos AI model to some companies and government agencies

Anthropic has limited access to two of its newest AI models, including Mythos 5, after the Trump administration issued an export control directive, while also allowing the model to be released to some companies and government agencies, according to CNBC. The reporting describes the action as tied to U.S. national security authorities’ assessment and the resulting compliance steps Anthropic took for controlled distribution.

CNBC reports that Anthropic disabled access to its Fable 5 and Mythos 5 models as part of the government directive’s implementation. The company then permitted release of Mythos 5 to a narrower set of customers, including select private firms and government agencies, aligning its rollout with export-control requirements cited as related to national security.

The CNBC account frames the change as an example of how U.S. export controls can affect the availability of frontier-model capabilities, even for providers already operating an API and customer access framework. Under the described approach, access is not blanket but instead tied to which entities are permitted to receive the model under the directive’s conditions.

The report, published June 26 and dated June 27, describes the trigger as the U.S. government’s export control directive rather than a voluntary change in Anthropic’s product decisions. It also characterizes the government rationale as originating from “national security authorities,” which the report says were involved in the compliance determination that led to the access restrictions.

Because the directive’s full text, the specific export classification, and the precise conditions for approved recipients were not included in the summary reported by CNBC, it is not possible to confirm in this account the exact criteria used to determine eligibility. The report indicates that some companies and government agencies were included in the permitted group, implying the directive operates through a review or authorization process rather than an all-or-nothing ban.

In practical terms, the next steps for affected customers likely depend on whatever authorization, contractual, and technical safeguards were required by the export-control process. For Anthropic, the described approach means continuing to manage two separate tracks: restricted access for non-approved customers and permitted availability for entities cleared under the government framework.

For the U.S. government, the decision underscores that national security considerations can translate into specific limits on the distribution of advanced AI systems. For the AI industry and downstream users, it highlights that model access may turn on regulatory treatment and entity-by-entity permissions, with potential spillover effects for research timelines, product roadmaps, and government procurement planning.