The U.S. government has allowed the AI company Anthropic to redeploy its most powerful cybersecurity model to a limited group of American organizations, ending a suspension it had imposed roughly two weeks earlier. Commerce Secretary Howard Lutnick authorized the model, Claude Mythos 5, for use by more than 100 U.S. companies and government agencies, TechCrunch reported.
What was approved
"I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model," Lutnick said in the directive, according to TechCrunch. Anthropic describes Mythos 5 as its "strongest cybersecurity model" — a system built to help defenders find and patch software vulnerabilities, though the same capabilities can in principle be turned to offense.
The clearance is narrow by design. It covers a defined set of organizations rather than the open market, and it reverses an earlier rule that had barred non-American employees at those organizations — and Anthropic's own international staff — from using the model. They are now permitted access under the new terms.
Why it had been pulled
The government suspended the model around June 12, roughly two weeks before the new authorization. The trigger, as reported, was a demonstration by security researchers that the model's safety guardrails — the built-in limits meant to stop it from assisting harmful tasks — could be bypassed relatively easily. Officials were also concerned, according to Semafor's reporting relayed by Investing.com, that a model with strong cybersecurity skills could be exploited by the military or intelligence services of adversaries such as China or Russia if it spread too widely.
That logic mirrors how Washington treats advanced semiconductors: so-called frontier AI models — the most capable systems at the leading edge of the technology — are increasingly handled as strategically sensitive, with access managed rather than left fully open.
What it does not cover
The directive did not address every Anthropic system. The company's Fable 5 model, which TechCrunch reported was widely released a couple of days before the ban and shipped with more safety protections, was not part of Lutnick's authorization. In other words, the government acted specifically on the high-capability Mythos 5, leaving other releases on a separate track.
Why it matters
For Anthropic, the decision restores access to a flagship product, but only inside a government-defined perimeter — an unusual arrangement in which a cabinet secretary effectively serves as the gatekeeper for who may use a private company's software. The selection of which organizations qualify, and on what criteria, has drawn scrutiny, with critics noting the process has not been spelled out publicly.
For the wider AI industry, the episode is a marker of how fast the U.S. is moving to treat the most capable models as controlled technology, akin to advanced chips or encryption. Whether the selective-access model applied to Mythos 5 becomes a template for future frontier systems, or stays a one-off, will shape how AI firms build and release their strongest products. Anthropic is a competitor to other large AI developers, and the framework set here could apply to all of them.
