CISA deploys Anthropic Mythos to hunt bugs in government code amid White House tensions

CISA Taps Anthropic Mythos to Scan Government Code Repositories

The Cybersecurity and Infrastructure Security Agency is actively using Anthropic Mythos to audit government software, marking a significant expansion of AI-powered cybersecurity tools within federal systems 1

1

. Three sources familiar with the matter confirmed that CISA is deploying the AI model Mythos to scan government code repositories for bugs that could leave critical infrastructure vulnerable to foreign spies and cybercriminals 2

2

. The US cyber agency's Attack Surface Evaluation team, a specialized group that conducts digital security assessments and hacking exercises across government, is leading the initiative .

Government Code Audit Uncovers Large Number of Security Vulnerabilities

Two sources indicated that the audits conducted by CISA have already uncovered a large number of vulnerabilities, though specific details about the nature or severity of the bugs remain undisclosed 2

2

. The exact scope of how much government code the team has examined remains unclear. This deployment to audit government software for vulnerabilities demonstrates growing federal confidence in AI-powered cybersecurity tools, even as Anthropic faces scrutiny over AI policy and national security concerns. The AI model has been described as extremely capable at finding and exploiting cybersecurity vulnerabilities, making it particularly valuable for defensive operations .

Anthropic's Turbulent Relationship with Federal Agencies

The San Francisco-based company, which has confidentially filed for a US initial public offering, has navigated a tumultuous relationship with the US government 1

1

. Relations hit a low point in February when Anthropic refused to remove AI safeguards that prevented its technology from being used for autonomous weapons or domestic surveillance. This stance prompted the Pentagon to issue a formal supply-chain risk designation—a label typically reserved for foreign companies suspected of facilitating espionage 2

2

. A judge blocked the extraordinary blacklisting in March, and tensions have eased following the private release of Mythos to select government agencies.

NSA Adoption and White House Tensions Over AI Safeguards

The NSA has been using Mythos since April despite the earlier blacklist, with analysts testing the model in classified settings and expressing strong approval of its capabilities . However, when Anthropic released a public version called Fable with cybersecurity safeguards, the White House demanded restrictions on foreign access. This triggered a global shutdown of the model that was only lifted last week 2

2

. The ongoing standoff highlights the complex balance between maintaining AI safeguards and meeting national security demands. As federal agencies increasingly adopt advanced AI models for critical cybersecurity functions, questions about access controls, AI policy frameworks, and the role of private companies in protecting government infrastructure will likely intensify.