Anthropic said on Friday that the U.S. government has allowed the company to restore access to its advanced Claude Mythos 5 artificial intelligence model for a limited group of trusted U.S. organizations, partially reversing restrictions imposed over national security concerns.
More than 100 companies and institutions, including many Fortune 500 companies, will now have access to Mythos 5 under the new directive, according to a source familiar with the decision. The source declined to be identified due to the sensitivity of the matter.
The decision follows a June 12 export control order that led Anthropic to suspend access to its most advanced AI models, Mythos 5 and Fable 5, for all users. The restrictions were introduced as the Trump administration increased oversight of advanced AI systems over concerns that powerful models could be misused by military or intelligence organizations in China, Russia, and other countries identified as security risks.
Anthropic said in a statement that the government had approved the redeployment of Mythos 5 to U.S. organizations responsible for operating and defending critical infrastructure.
“Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure,” the company said.
Anthropic added that it was restoring access quickly and continuing discussions with the government to expand availability of Mythos 5 and allow Fable 5 to return for broader use.
The government’s process for determining which organizations can access the model has drawn criticism from transparency and technology policy groups. John Coleman, legislative counsel for the Foundation for Individual Rights and Expression, questioned how companies are selected and why other organizations remain excluded.
Coleman said the process gives the government significant control over access to advanced AI systems and raises concerns about transparency and legal oversight.
OpenAI chief executive Sam Altman also raised concerns about government involvement in selecting AI model users. In a post on X, Altman said safety testing was reasonable but questioned whether the government should determine which customers receive access to leading AI systems.
Experts have warned that advanced AI models could increase the ability to conduct sophisticated cyberattacks if misused, particularly against industries such as banking that rely on complex and aging technology infrastructure.
A letter from U.S. Commerce Secretary Howard Lutnick to Anthropic said the company had made “significant progress” in working with the government to address risks associated with the company’s advanced models. However, it remained unclear what specific safeguards had been implemented.
Anthropic previously said it believed the government had identified a possible method of bypassing, or “jailbreaking,” safeguards designed to prevent Fable 5 from being used to identify software vulnerabilities.
Lutnick said export licenses would no longer be required for approved companies and their non-U.S. citizen employees to access Mythos 5. Licensing restrictions would remain for companies that are not included on the approved list.
The source familiar with the directive said many approved organizations are part of Anthropic’s Project Glasswing, which includes around 100 technology companies and institutions. The government is also moving toward allowing Anthropic to release Fable 5, although no timeline has been provided.
Mythos 5 and Fable 5 use the same underlying AI model, but Fable 5 is intended for wider public availability while Mythos 5 operates with certain safeguards adjusted for specialized use.
Anthropic’s relationship with the U.S. government has faced tensions after the company declined to allow its AI systems to be used by the military for domestic surveillance or fully autonomous weapons systems. The government later placed Anthropic on a national security blacklist.
The restrictions on Anthropic and OpenAI followed President Donald Trump’s signing of an executive order establishing a voluntary framework requiring developers of certain advanced AI models to provide them to the government for up to 30 days before releasing them to trusted partners.
Kate Koren, an analyst at the Center for Strategic and International Studies and a former Commerce Department official, said the latest order was a temporary measure but left unresolved broader questions about how U.S. companies can release updated AI models at scale.
Koren said delays in establishing a long-term system for AI model releases could affect competition between U.S. companies and China.
