US Government Authorizes Anthropic to Restore Mythos 5 to 100+ Institutions

The US government has authorized Anthropic to redeploy its Mythos 5 cybersecurity AI model to more than 100 US institutions, including major corporations and government agencies, according to a company announcement and reporting by Semafor.

Commerce Secretary Howard Lutnick formally approved the redeployment in a letter to Anthropic, stating: "I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model." The letter also noted that Anthropic "has committed to work with the US government on protocols and standards and releases" for its models.

Two-Week Suspension

Anthropic suspended all customer access to Mythos 5 and Fable 5 on June 12, 2025, following a directive from the US government. The government requested Anthropic block access for all foreign nationals, including those working for Anthropic within the US.

According to reports, the directive came after Amazon and other companies warned the government that the models could be jailbroken and used maliciously. US authorities were also concerned that a China-linked group had gained access to Mythos 5.

Fable 5, designed to bring many of Mythos 5's capabilities to the public, remains suspended. Anthropic stated it is in talks with the government to restore access but provided no timeline.

Daily Negotiations

Lutnick's letter revealed that Anthropic made significant progress through daily talks with the government over the past two weeks since the suspension began.

Anthropic stated it is "redeploying Mythos 5 quickly and continuing to work with the government to expand access even further" beyond the initial 100+ approved organizations that operate and defend critical infrastructure.

What This Means

This marks the first known instance of the US government directly ordering a major AI company to suspend access to a deployed model over security concerns, then conditionally reinstating it with restrictions. The incident establishes a precedent for government intervention in AI model deployment when national security concerns arise, particularly around cybersecurity-focused models that could be weaponized. The requirement for "trusted partners" suggests a new category of restricted AI models that may require government vetting for access, similar to export controls on sensitive technology.