Anthropic regains approval to deploy Mythos AI to trusted US organisations
'Significant progress' was made during nearly two weeks of discussions between the government and Anthropic following the restrictions
The US government has partially lifted restrictions on Anthropic's most powerful AI model, allowing a select group of trusted American organisations to regain access after a two-week suspension prompted by national security concerns.
Anthropic said on Friday it had been informed by the government that its Claude Mythos 5 model could once again be deployed to organisations responsible for operating and protecting critical infrastructure in the United States.
The San Francisco-based AI firm said it was restoring access immediately and would continue working with officials to broaden availability of Mythos 5 and eventually return its Fable 5 model to general use.
The Trump administration's June 12th export control order had forced Anthropic to abruptly suspend access to both Mythos 5 and Fable 5 worldwide over concerns the systems could be exploited for malicious purposes.
A source familiar with the matter told Reuters that more than 100 companies and institutions will now be eligible to use Mythos 5.
Many of them are members of Anthropic's Project Glasswing, an initiative that includes major technology companies and other organisations working on cyber defence and critical infrastructure protection.
Sufficient safeguards now in place
In a letter sent to Anthropic on Friday, US Commerce Secretary Howard Lutnick said he had determined that sufficient safeguards were now in place to permit access for approved organisations.
Lutnick cited "significant progress" made during nearly two weeks of discussions between the government and Anthropic following the restrictions.
Under the revised policy, approved companies will no longer require export licences to allow authorised employees, including non-US citizens, to use Mythos 5. Anthropic's own non-US employees working on the project will also be exempt from the licensing requirement.
However, organisations outside the approved group will continue to face export restrictions.
In a post on X, Anthropic said it had worked closely with officials since the restrictions were introduced and welcomed the restoration of access for organisations defending critical infrastructure.
While the government's letter focused solely on Mythos 5, it did not address the future of Fable 5, Anthropic's less powerful consumer-oriented model.
According to Reuters, officials are also moving towards allowing its return, although no timetable has been agreed.
Anthropic joins effort to protect open source software
The US government's decision to ease restrictions on Anthropic's flagship AI model came in the same week that the Linux Foundation unveiled a new industry initiative, joined by 20 founding organisations, to strengthen the security of critical open source software.
The programme, called Akrites, brings together Anthropic, Google, Microsoft, AWS and other major technology organisations to coordinate the discovery and disclosure of software vulnerabilities.
The group will establish a shared Security Incident Response Team (SIRT) and adopt a common Coordinated Vulnerability Disclosure (CVD) process to streamline how security flaws are reported, fixed and disclosed.
Akrites aims to replace what organisers describe as a fragmented security response, in which multiple organisations often investigate the same vulnerabilities independently, generating duplicate reports or competing software patches that place additional strain on volunteer maintainers.
Instead, the initiative will provide a single coordination point where participating organisations can jointly investigate vulnerabilities, develop fixes and manage responsible disclosure before details become public.
The shared incident response team will also work with operators of critical infrastructure to help deploy patches before attackers can exploit newly discovered flaws.
Confidentiality will remain central to the process. Security fixes will continue to be contributed through projects' existing maintainers wherever possible.
The initiative also plans to coordinate with government agencies to improve collaboration between public and private sector defenders during major security incidents.
Jason Clinton, Anthropic's deputy chief information security officer, said the rapid advances in AI had outpaced traditional approaches to vulnerability disclosure.
"Open source projects collectively underpin much of the internet, and the existing model for coordinated disclosure has been outpaced by how quickly AI can now find vulnerabilities," he said.
"Getting ahead of that requires the industry to coordinate on findings and get fixes upstream before they're disclosed and exploited. Efforts like Akrites drive this level of coordination at the scale and speed this moment requires."