The Donald Trump administration is close to allowing Anthropic to restore access to the powerful Fable 5 model, which has remained unavailable for 15 days because of government security concerns, according to sources familiar with the situation.
According to one source familiar with the matter, restrictions on Fable 5 could be lifted as early as next week. A second source said talks will continue over the weekend, and Anthropic expects to restore access to the model soon.
For developers and technically sophisticated users, the shutdown of Fable 5 was an unprecedented event: one of the most powerful models, already available to users, was effectively taken out of service because of government intervention.
Progress toward unblocking Fable 5 points to an easing of tensions after a four-month standoff between the Trump administration and Anthropic.
Another sign of de-escalation came Friday, when the U.S. Department of Commerce allowed Anthropic to restore access to Mythos 5—the company’s most powerful cybersecurity model—for a limited number of trusted users. Mythos 5 is equipped with safeguards intended to prevent its use for cyberattacks or biological terrorism and was never freely available.
Commerce Secretary Howard Lutnick said in a letter to Anthropic that the company had “worked with the U.S. government to address the risks” associated with Mythos 5 and Fable 5. Those efforts, he said, had “led to significant progress.” Anthropic also pledged to cooperate with authorities on protocols, standards and the release of new models.
The return of Fable 5 is being awaited by users who praised its ability for deep analysis and fast, complex coding. Developers described the leap in the model’s capabilities as impressive, and new models—especially open ones—are now being compared specifically with Fable 5.
The final decision still has to be approved by the Pentagon and the National Security Agency, so the outcome remains uncertain. However, other agencies have already concluded that Fable 5 can safely be returned to public access.
Lutnick himself and Treasury Secretary Scott Bessent helped soften the conflict between the administration and Anthropic.
One administration official told the outlet that Anthropic had “worked positively with the government.” That is a notable contrast with an earlier harsh statement by Defense Secretary Pete Hegseth, who called Anthropic “a national security risk in the supply chain” after he and the company’s head, Dario Amodei, failed to agree on the terms for the Pentagon’s use of Claude.
Anthropic presented Fable 5 as the most powerful model ever released to the general public. Before access was cut off, Every’s Vibe Check newsletter called it “the best coding model in the world”—just three days after launch.
In early tests cited by Anthropic, the payments company Stripe used Fable 5 to rework a 50 million-line codebase in a single day—a task that would have taken engineers more than two months to do manually.
When access disappeared on June 12, developers’ automated tasks stopped, and companies began urgently switching to competing solutions, including cheaper Chinese models.
Anthropic initially provided Fable 5 at no additional cost in several paid Claude plans until June 22, giving users a short period to test the model’s capabilities. It is not yet known whether subscribers will receive the promised free access after Fable 5 is restored—or whether the model will return with an additional fee or identity verification.
Anthropic and OpenAI are now pressing the administration to formalize the procedure for reviewing new models. Such a mechanism was envisioned in Trump’s June 2 executive order, which created a framework for voluntary government evaluation of the most powerful new AI models.
The companies are dissatisfied with the current approach, under which each decision is made separately. After access to Fable 5 and Mythos 5 was suspended on June 12, Anthropic called for “a legislatively established process—transparent, fair, understandable and based on technical facts.” The company said the administration’s decision to restrict the two models did not meet those principles.
OpenAI, which on Friday was allowed to begin a limited preview of GPT-5.6, also said that this kind of government access regime should not become the long-term norm. According to the company, it limits access to the best tools for users, developers, businesses, cyberdefense specialists and international partners.