The U.S. Government is Using Anthropic AI to Scan its Own Flaws

Governments are increasingly turning to artificial intelligence to scan millions of lines of software code for hidden vulnerabilities as cyberattacks from state-backed hackers and criminal groups grow more sophisticated. The latest example comes from the United States, where the country’s top cybersecurity agency is quietly using Anthropic’s AI model to audit government code for security flaws despite the company’s recent tensions with the White House and the Pentagon.

The federal government’s top cybersecurity agency is quietly using an advanced artificial intelligence model from Anthropic to comb through government software for security vulnerabilities. The initiative reflects a broader shift in Washington, where AI systems are increasingly being deployed to strengthen national cyber defenses even as the government continues to wrestle with questions about how such technologies should be regulated and secured.

It shows Anthropic’s unusual position in Washington. The San Francisco AI company has become one of the government’s most sought-after providers of advanced cybersecurity tools despite months of friction with the White House and the Pentagon over restrictions built into its models. The same company that was recently at odds with U.S. officials over military and surveillance uses of artificial intelligence is now helping identify weaknesses in government software that could be exploited by foreign intelligence agencies or criminal hackers.

The Cybersecurity and Infrastructure Security Agency, or CISA, is using Anthropic’s AI model, Mythos, to scan government code repositories for programming flaws. The work is being carried out by CISA’s Attack Surface Evaluation team, a unit that conducts cybersecurity assessments and simulated hacking exercises across federal agencies.

The audits have already uncovered a significant number of software vulnerabilities, although they declined to provide details. It was not immediately clear how much government code has been reviewed or how severe the identified flaws are.

The initiative underscores the growing role of generative AI in cybersecurity, where companies increasingly market large language models as tools capable of identifying software bugs, reviewing millions of lines of code and even discovering previously unknown vulnerabilities far faster than human analysts.

For Anthropic, the project marks another turn in an often complicated relationship with the U.S. government.

Earlier this year, the company resisted pressure to weaken safeguards that prevented its AI systems from being used for autonomous weapons or domestic surveillance, prompting the Pentagon to designate Anthropic as a supply chain risk. A federal judge later blocked that designation.

Relations have improved since Anthropic privately released Mythos, a model designed specifically for cybersecurity applications. The National Security Agency has been using the system since at least April, according to Axios, and The New York Times reported last month that NSA analysts testing the model in classified environments were impressed by its capabilities.

Anthropic’s cybersecurity models have also become entangled in the geopolitical contest over artificial intelligence. After the company released a public version of Mythos known as Fable, the White House directed Anthropic to block access by foreign users, leading to a temporary worldwide shutdown before service resumed last week.

This shows a broader tension confronting AI companies and governments alike. As advanced models become increasingly valuable for cyber defense, they are also viewed as potential tools for offensive hacking, espionage and intellectual property theft, forcing companies to balance commercial ambitions with national security concerns.

Stay ahead of the stories shaping our world. Subscribe to Impact Newswire for timely, curated insights on global tech, business, and innovation all in one place.

Dive deeper into the future with the Cause Effect 4.0 Podcast, where we explore the ideas, trends, and technologies driving the global AI conversation.

Got a story to share? Pitch it to us at info@impactnews-wire.com and reach the right audience worldwide