Trump Administration Lifts Export Restrictions on Claude Fable 5

The Trump administration has lifted export restrictions on Anthropic’s Claude Fable 5 model. Access for customers is expected to return on Wednesday, Anthropic said.

The decision restores public access to a powerful Mythos-class model that was shut off 18 days ago over security concerns.

Last week, the Trump administration had already allowed Anthropic to restore access to Mythos 5 for a limited group of government-approved organizations. Sources also said Fable 5 could return as early as this week.

The situation shows that the U.S. government’s role in regulating and reviewing advanced AI models before their release remains uncertain. As a result, developers are effectively facing a manual and ad hoc regulatory regime.

Last week, OpenAI released GPT-5.6 only to a small group of approved customers after a request from the U.S. government. Anthropic is similarly opening Mythos 5 to a group of customers coordinated with the authorities.

U.S. Commerce Secretary Howard Lutnick said on X that his department had “worked closely with Anthropic to analyze and approve Fable 5, ensure alignment across the U.S. government and strengthen America’s leadership in AI.”

“We are grateful to our users for their patience and to everyone who worked with us on redeploying the models,” Anthropic said.

White House Chief of Staff Susie Wiles wrote on X that the state and the private sector had “worked together in a way we have never seen before,” and that the America First policy created on that basis is “unprecedented.”

Anthropic says it worked with the government to introduce a new safeguard that blocks, in 99% of cases, the jailbreak that had raised officials’ concerns.

In the remaining cases, according to the company, Fable’s responses contained only previously discovered or already fixed vulnerabilities. At the same time, the new protection may lead the model to more often flag harmless requests as suspicious.

According to Anthropic, the U.S. Commerce Department’s Center for AI Standards and Innovation also tested the model’s previous and new safeguards.

U.S. officials, American allies and leading AI companies are increasingly concerned that advanced models could be used to automate sophisticated cyberattacks or accelerate the development of biological weapons.

At the same time, China is moving closer to creating its own Mythos competitor and continues to release open-weight models comparable to systems such as Claude Opus 4.8.

The next step for the Trump administration will be to create standardized criteria for assessing the security risks of new AI models. Under a recent executive order, this is to be done by August.