The United States government has removed the licensing requirement that had prevented Anthropic from distributing its Mythos and Fable artificial intelligence models to users outside the country. The restriction had effectively shut down public access to what many in the industry regard as the most capable AI models released to date.
Anthropic confirmed that it would begin restoring access to both models starting Wednesday, July 1, following weeks of negotiations between the company and federal officials.
How the Restrictions Came About
The situation began on June 12, when the U.S. government placed Mythos and Fable on its list of export-controlled technologies. That designation meant the models could no longer be made accessible to foreign nationals without first obtaining special government approval. Complying with that requirement at any meaningful scale proved unworkable, and Anthropic was left with little choice but to suspend public access to the models entirely.
The move drew immediate scrutiny from cybersecurity professionals and policy observers, many of whom questioned whether the restrictions served any genuine national security purpose. Critics noted that Anthropic had already made voluntary public commitments to responsible deployment and security monitoring well before the export rule was introduced, which led some analysts to conclude that the ban functioned more as a political instrument than a protective measure.
Some observers interpreted the restrictions as a way for the Trump administration to apply pressure on Anthropic, whose executives had at times been publicly critical of how the government and certain political figures might seek to use advanced AI technology.
The Agreement That Ended the Standoff
Following the resolution of those talks, Secretary of Commerce Howard Lutnick announced that Anthropic had committed to a set of obligations as a condition of lifting the export controls. According to Lutnick, the company agreed to proactively identify and address security risks tied to the models, to collaborate closely with the U.S. government on protocols, standards, and release procedures for Mythos, Fable, and any future models, and to notify federal authorities of any detected malicious activity involving the technology.
The commitments largely mirror what Anthropic had already pledged to do on a voluntary basis months earlier, a fact that reinforced skepticism among security experts about the original rationale for the export ban.
Background on the Models
Mythos was initially made available in April, but only to a carefully selected group of organizations. That limited rollout was designed to address concerns about the model's capacity to identify and exploit software vulnerabilities. A variant called Fable followed in June, released more broadly to the public and built with additional security guardrails intended to reduce potential misuse.
The pressure on Washington to ease its stance on Anthropic intensified as AI developers in Asia began releasing models with capabilities approaching those of Mythos. Among the notable examples were systems known as Fugu and Tulongfeng, which signaled that American dominance in frontier AI development could not be taken for granted. That competitive dynamic gave the U.S. government added incentive to allow Anthropic's most advanced models to reach global markets.
Selective Access and Industry-Wide Uncertainty
In the week before the full restrictions were lifted, Lutnick cleared Mythos for distribution to a limited pool of customers who had received White House approval. A similar arrangement was applied to OpenAI's latest models, which were made available to a set of organizations vetted by the Trump administration rather than being released to the general public.
The broader pattern of decision-making around AI policy has created significant uncertainty across the technology industry. Companies developing or deploying advanced AI systems have struggled to anticipate what rules will govern future model releases, given the inconsistent signals coming from the administration.
An executive order issued in June suggested that the government was interested in reviewing AI models before they are made public. That order drew criticism from prominent voices in the policy community, including Dean W. Ball, an influential analyst who recently moved into a policy role at OpenAI. Ball and others argued that such a review process, if implemented broadly, could introduce delays and regulatory friction that would disadvantage American AI developers relative to their international competitors.
The episode surrounding Mythos and Fable has highlighted the tension between national security considerations and the competitive realities of the global AI race, a tension that policymakers and the industry will likely continue to navigate as more powerful models emerge in the months ahead.
