New era of cyber threats

House Homeland gets live demonstration of Anthropic’s Mythos model

The briefing “made clear that we have entered a new era of cyber threats,” said Rep. James Walkinshaw.

By Dana Nickel and Maggie Miller

Representatives from Anthropic on Wednesday gave a live demonstration of the company’s highly advanced AI model, Claude Mythos, to members of the House Homeland Security Committee.

The closed-door briefing “made clear that we have entered a new era of cyber threats,” said Rep. James Walkinshaw (D-Va.), a member of the panel, in a statement.

Lawmakers took the opportunity to raise concerns about the national security threat posed by frontier AI models, which are quickly outpacing some humans’ abilities to find and exploit cyber vulnerabilities. Mythos has so far only been tested by a small group of researchers, agencies and tech companies because of its unprecedented hacking power.

Committee Chair Andrew Garbarino (R-N.Y.) said that Anthropic “showed us how it works,” including how to use it “as a defensive tool.”

A spokesperson for Anthropic did not respond to a request for comment.

Still, he said, committee members remain most concerned about how adversaries or cybercriminals may co-opt the tool to launch cyberattacks at a scale unseen.

“When will China catch up?” Garbarino said. “Everybody’s concerned about us not using this now to defend ourselves when China … could only be six to 12 months behind us.”

Rep. Lou Correa (D-Calif.) said Anthropic representatives used Mythos to demonstrate “how to break into my bank account.”

“With the AI, they were able to analyze the weaknesses in the code of certain areas, and using that weakness, they would be able to break in and steal all of the above,” he added.

Lawmakers also discussed expanding access to highly advanced models, such as Mythos or OpenAI’s recently announced GPT-5.5-Cyber model, to better support vulnerable U.S. networks.

Walkinshaw said that “safely and securely putting advanced AI cybersecurity capabilities in the hands of state and local governments, critical infrastructure owners and other high-risk entities must be a national priority.”

According to Garbarino, lawmakers are pushing for CISA to gain access to Mythos, as other federal agencies reportedly have, to test the model on its networks.

“After seeing it today, I think it would be great if we could use it in our government service just to run a check [on vulnerabilities], and have CISA use it as a defense tool,” he added.

A spokesperson for CISA did not immediately respond to a request for comment on whether the agency has tested the model.

Other key lawmakers told POLITICO they are interested in seeing Mythos’ hacking capabilities in action, including House Armed Services Chair Mike Rogers (R-Ala.), Senate Homeland Security ranking member Gary Peters (D-Mich.) and Senate Armed Services cyber Chair Mike Rounds (R-S.D.).

Senate Intelligence ranking member Mark Warner (D-Va.) said he and Senate Intelligence Chair Tom Cotton (R-Ark.) were briefed on Mythos by Gen. Joshua Rudd, head of both Cyber Command and the National Security Agency, but declined to give further details.

“This is such a powerful tool, and there will be many others that will come after,” he said.

A person familiar with the ongoing meetings, granted anonymity to discuss the closed-door sessions, confirmed that Anthropic will be back on the Hill this week to brief more lawmakers on Mythos.

“The more members that see it, the better,” Garbarino said.