Pete Hegseth Called Anthropic a ‘National Security Risk.’ Now CISA Is Using It.

National Interest  |  Peter Suciu

The United States Cybersecurity and Infrastructure Security Agency (CISA) has deployed Anthropic’s restricted Mythos artificial intelligence model to audit federal software code for critical vulnerabilities. This deployment directly challenges the Pentagon's previous designation of the developer as a national security risk, highlighting deep policy inconsistencies within the American government regarding advanced technology adoption.

Historically, federal agencies have struggled to secure legacy systems running outdated code, but the integration of large language models allows rapid scanning of vast codebases that previously required months of manual human penetration testing. However, cybersecurity experts warn that the platform may generate high volumes of false positives or unactionable data, requiring significant human oversight to prioritize actual exploitability. Furthermore, because developers increasingly use generative tools that introduce fresh coding flaws, the agency must establish secure-generation standards and expand scanning access to critical infrastructure operators to prevent continuous vulnerability cycles.

Comment

State adoption of offensive-capable artificial intelligence tools creates a dangerous dual-use dilemma. National security agencies must establish clear boundaries between defence-oriented auditing and autonomous exploitation. Unregulated deployment of these platforms will inevitably accelerate the global cyber arms race.

To read in details click here