CISA is running Anthropic's Mythos AI model against federal code repositories to catch vulnerabilities before foreign intelligence services and criminal groups find them, three sources told Reuters.

 

What happened

CISA's Attack Surface Evaluation team, the unit that runs security assessments and simulated attacks across the federal government, is using Mythos to scan government code for bugs. Two sources told Reuters the audits have already surfaced a large number of vulnerabilities, though the scope, which agencies were covered, and how serious the flaws are remain undisclosed. Neither CISA nor Anthropic commented on the record. Mythos is Anthropic's most capable model and isn't available through a standard subscription. Anthropic released it privately to select government partners, describing it as exceptionally capable at finding and exploiting security vulnerabilities. The NSA has reportedly used the same model since at least April.

 

The backstory

Anthropic's relationship with the U.S. government turned hostile in February when the company refused to strip safeguards that blocked Mythos from being used for autonomous weapons or domestic surveillance. In response, the Pentagon designated Anthropic a supply-chain security risk, a label previously reserved for foreign companies suspected of facilitating espionage. A federal judge blocked the blacklisting in March. When Anthropic launched Fable in early June, a public version of Mythos with added cybersecurity safeguards, the White House demanded the company ban foreign nationals from running it. That demand triggered a temporary global shutdown of the model, which was lifted only last week.

Read also: Commerce department lifts export controls on two Anthropic AI models

 

What was said

On the February dispute that triggered the Pentagon blacklist, an Anthropic spokesperson told CNBC that seeking judicial review "does not change our longstanding commitment to harnessing AI to protect our national security," calling the lawsuit "a necessary step to protect our business, our customers, and our partners," and adding that Anthropic would "continue to pursue every path toward resolution, including dialogue with the government."

Anthropic CFO Krishna Rao quantified the financial stakes of the standoff in a court filing, warning that "the government's actions could reduce Anthropic's 2026 revenue by multiple billions of dollars." The figure is part of the backdrop against which CISA's decision to actively deploy Mythos now reads as a reversal of fortune for the company.

When a federal judge blocked the Pentagon's blacklist in March, she rejected the government's underlying theory directly, writing that nothing in the law supports the notion that an American company can be branded "a potential adversary and saboteur of the US for expressing disagreement with the government."

 

Why it matters

The capability that got Anthropic blacklisted as a supply-chain risk in February (an AI powerful enough to autonomously find and exploit software vulnerabilities) is the same capability CISA and the NSA are now relying on to defend federal systems. Meanwhile, the public-facing version of that same technology, Fable, triggered an immediate shutdown over foreign-access concerns. That approval for government-only deployment and regulatory alarm for public deployment shows that the government's real concern isn't the capability itself but who controls it.

 

The bottom line

Mythos finding flaws in federal systems "in hours" rather than weeks is a preview of how fast AI-driven vulnerability discovery is moving on both the offensive and defensive sides of cybersecurity. Agencies now have a strong incentive to expand these programs, but the undisclosed scope and severity of what's been found leaves a question about how exposed government systems were in the first place and how quickly adversaries could get similar capabilities.

 

FAQs

What's the difference between Mythos and Fable?

Mythos is Anthropic's most capable AI model, restricted to select government and vetted partners, while Fable is a public-facing version with added cybersecurity safeguards.

 

Can ordinary users access Mythos?

No, Mythos is only available to select government partners and vetted organizations, not through a standard public subscription.

 

What is a supply-chain risk designation?

It's a formal U.S. government classification signaling that a company is considered a potential security threat, historically applied almost exclusively to foreign firms suspected of aiding espionage.