Anthropic Mythos 5 cleared for wider US use, with limits

The Trump administration on Friday let Anthropic return Mythos 5 to a limited group of domestic users, pulling back from a two-week clampdown that had put one of the US AI sector’s sharpest policy fights on display. More than 100 trusted US organizations, companies and institutions can use the model. Fable 5 is still under the tougher curb.

This is a narrow opening, not a product launch. Officials have moved Mythos 5 from a freeze to a selective-release system, keeping the technology inside an approved circle while preserving the national-security case for intervention. The shift also avoids sidelining a domestic AI company as rivals such as OpenAI compete for enterprise customers. Bloomberg and CNBC reported that the pool includes selected companies and government bodies, not the broader commercial market.

For companies trying to plan around the rules, that distinction is the story. A ban is easy to understand. A trusted-partners carve-out is discretionary, and it leaves applicants guessing where the next line will be drawn.

Anthropic still gets a commercial reprieve. Mythos 5 is the company’s strongest cybersecurity-focused system, and a blanket block risked turning a policy dispute into a competitive handicap with critical-infrastructure customers. The partial reopening gives the company some room to serve that market while letting officials say they are backing US AI leadership without dropping guardrails altogether.

In a statement carried by Reuters, Anthropic said Mythos 5 could again be deployed to groups that run and defend essential systems.

“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.”

Anthropic, as cited by Reuters

The company did not disclose the full list of eligible users. By stressing critical infrastructure, though, it narrowed the opening to operators whose work already sits close to Washington’s security case. The administration can still say the model has not been released to the wider market.

The episode began with a June 12 export-control order that suspended access to the model. Two weeks later, Commerce Secretary Howard Lutnick said safeguards were sufficient for a narrower domestic rollout.

“I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model.”

Howard Lutnick, as cited by CNBC

A selective reversal

CNBC said the approved group included government agencies as well as private-sector entities, while Bloomberg reported that Mythos 5 had been cleared for wider use without the same relief for Fable 5. Keeping one model restricted while reopening another gives officials more ways to apply pressure. For AI companies, it points to permissions being set case by case.

The reversal leaves Washington with a credibility problem. If Mythos 5 was too risky for release earlier this month, officials now have to explain what changed in two weeks beyond the political and commercial cost of holding the line. Semafor reported that administration officials framed the move as a way to keep the US in front in AI while preserving security controls, language that shows industrial policy and export policy being mixed in real time.

Benno Kass, a White House special adviser on AI and crypto, made that balancing act explicit in remarks reported by Semafor.

“In just two weeks, we have worked diligently to ensure America remains the global leader in AI while safeguarding our security.”

Benno Kass, as cited by Semafor

For investors, customers and rivals, the message is practical and uncomfortable: access to top-tier AI systems may depend less on a public standard than on whether an institution fits inside a trusted-user framework. Anthropic gets relief for now. The next restriction will be harder to price if the winners and losers are first sorted in Washington, then explained to the market.