The United States’ top cyber defence agency is using Anthropic’s advanced artificial intelligence model Mythos to identify vulnerabilities in government software, highlighting Washington’s growing confidence in AI-powered cybersecurity despite recent tensions between the AI company and the White House.
According to people familiar with the matter, the Cybersecurity and Infrastructure Security Agency (CISA) has deployed Mythos to scan government code repositories for software flaws that could potentially be exploited by foreign intelligence agencies or cybercriminals.
The work is being carried out by CISA’s Attack Surface Evaluation Team, a specialised unit responsible for conducting cybersecurity assessments and ethical hacking exercises across US government systems.
Sources told Reuters that the AI-driven audits have already uncovered numerous software vulnerabilities, although the nature and severity of the flaws remain undisclosed. It is also unclear how much government code has been reviewed so far.
Neither Anthropic nor CISA has publicly commented on the initiative.
Rocky Ties With Washington
The reported collaboration comes despite a turbulent relationship between Anthropic and the US government earlier this year.
In February, the San Francisco-based AI company reportedly refused to remove safeguards that prevented its models from being used for autonomous weapons or domestic surveillance. The decision prompted the Pentagon to designate Anthropic as a supply-chain security risk a rare classification generally associated with companies suspected of facilitating espionage.
That designation was later blocked by a federal judge in March.
Relations have since improved following the private release of Mythos, an AI model specifically designed to identify and exploit cybersecurity vulnerabilities during defensive security testing.
Growing Government Adoption
Reports suggest the National Security Agency (NSA) has also been using Mythos since April despite the earlier Pentagon blacklist.
According to reports by Axios and The New York Times, NSA analysts tested Mythos in classified environments and were impressed by its cybersecurity capabilities.
However, Anthropic faced fresh controversy after releasing a public version of the model, Fable, which incorporated cybersecurity safeguards. The White House reportedly demanded restrictions preventing foreign users from accessing the model, leading to a temporary global shutdown that was lifted only last week.
The latest use of Mythos by CISA signals the US government’s increasing willingness to integrate advanced AI into national cybersecurity efforts as it seeks to strengthen defences against cyber espionage, software vulnerabilities and attacks targeting critical infrastructure.
The NSA and the White House have not commented on the reported deployment.