The U.S. government has approved Anthropic for a limited re-release of its Mythos 5 model. According to a letter from Commerce Secretary Howard Lutnick to Anthropic reviewed by NBC News, Mythos 5 access will be restored to around 100 organizations, including government agencies and private companies.

In the letter, Lutnick said the government was confident that Anthropic had put adequate safeguards in place to permit trusted partners to use the system. Anthropic said in a post that the model could be redeployed to organizations that operate and defend critical infrastructure.

The move follows an earlier U.S. action two weeks before that required Anthropic to shut off access to its two most capable systems—Mythos 5 and Fable 5—under export control authorities, with the government citing national security concerns. The reporting says the shutdown affected access that had previously been limited to trusted entities under Anthropic’s Project Glasswing.

Anthropic’s Mythos 5 was previously available only to a subset of trusted organizations, which included infrastructure providers and major banks. An early version had reportedly helped find thousands of cyber vulnerabilities and bugs. Fable 5, by contrast, had been available to the general public, and had stronger guardrails designed to reduce the risk of misuse across certain sensitive areas such as cyber and biology.

The approved limited re-release of Mythos 5 comes hours after Anthropic’s competitor, OpenAI, announced that it would release its latest models (the GPT-5.6 family) in phases per a federal government request. OpenAI’s CEO Sam Altman criticized the staggered approach as unfavorable to the company’s plans.

The reporting also says Anthropic will continue discussions with the U.S. government over the weekend as it seeks to restore access to Fable 5 as well, though timing was unclear.