The U.S. government is increasingly embracing artificial intelligence as a frontline cybersecurity tool, with the Cybersecurity and Infrastructure Security Agency (CISA) now using Anthropic’s AI model, Mythos, to identify security flaws in government software, according to sources cited by Reuters.
The move signals growing confidence in AI’s ability to strengthen national cyber defenses while highlighting a dramatic shift in Anthropic’s relationship with Washington. Just months ago, the AI company found itself at the center of a bitter dispute with the Trump administration that resulted in an unprecedented Pentagon blacklist.
According to three sources familiar with the initiative, CISA has deployed Mythos to examine government software code repositories for vulnerabilities that could be exploited by foreign intelligence agencies, state-sponsored hackers or cybercriminals.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
The work is being carried out by CISA’s Attack Surface Evaluation team, a specialist unit responsible for conducting cybersecurity assessments, penetration testing, and simulated hacking exercises across U.S. government systems.
Rather than relying solely on traditional vulnerability scanning tools, the agency is using Anthropic’s advanced AI model to comb through large volumes of source code, identify programming mistakes and flag weaknesses before attackers can exploit them.
Two of the sources said the AI-assisted reviews have already uncovered a significant number of software vulnerabilities, although they declined to provide details about the affected systems or the severity of the flaws.
Reuters said it could not independently determine how much government software has been reviewed or whether any of the discovered vulnerabilities were considered critical.
Neither Anthropic nor CISA publicly confirmed the programme.
Anthropic declined to answer questions about the initiative. A CISA representative had previously indicated the agency would determine whether it could share information about the project but did not respond to subsequent requests for comment.
AI’s Growing Role in Cybersecurity
Modern government software consists of millions of lines of code, making manual security reviews increasingly difficult and time-consuming. AI models trained to understand programming languages can analyze code at much greater speed, helping security teams identify vulnerabilities ranging from coding errors and insecure software configurations to authentication weaknesses and exploitable logic flaws.
The adoption of Mythos has raised eyebrows because it comes after months of strained relations between Anthropic and the U.S. government. The San Francisco-based company, which has confidentially filed for an initial public offering, became embroiled in a high-profile dispute with the Trump administration earlier this year over the safeguards built into its AI systems.
The disagreement escalated in February after Anthropic refused government requests to remove restrictions designed to prevent its AI models from being used for autonomous weapons or domestic surveillance. The company argued that those safeguards were essential to limiting dangerous uses of advanced artificial intelligence.
The refusal triggered an extraordinary response from the Pentagon, which designated Anthropic as a supply-chain risk. The classification is typically reserved for foreign companies suspected of enabling espionage or posing national security threats. Applying the designation to a leading U.S. artificial intelligence developer was widely viewed as an unprecedented move.
Anthropic challenged the decision in court, and in March, a federal judge blocked the blacklist from taking effect while legal proceedings continued. Since then, relations between the company and government agencies appear to have improved, driven largely by the cybersecurity capabilities of Mythos.
The AI model has been described as exceptionally effective at identifying and exploiting software vulnerabilities, making it valuable for defensive cyber operations.
But government interest in the technology extends well beyond CISA.
Earlier reports indicated that the National Security Agency (NSA), the United States’ premier signals intelligence agency, began using Mythos as early as April, even while the Pentagon’s supply-chain designation remained in place. According to previous media reports, NSA analysts tested the model in classified environments and concluded that it demonstrated impressive cybersecurity capabilities.
Those findings helped boost Anthropic’s reputation as one of the leading developers of AI systems tailored for cyber defense. The company later introduced a public version of Mythos called Fable, incorporating what it described as additional cybersecurity safeguards.
That launch, however, sparked another confrontation with the White House. Administration officials reportedly demanded that Anthropic prevent foreign users from accessing the model, citing concerns that hostile governments or cybercriminals could exploit its sophisticated vulnerability discovery capabilities.
The dispute resulted in Anthropic temporarily shutting down global access to Fable before restoring availability last week.
As cybersecurity concerns intensify, federal agencies face an expanding attack surface, especially due to the government’s systems becoming more interconnected and cyber threats growing increasingly sophisticated. AI-assisted code analysis offers the potential to identify vulnerabilities at a speed and scale that would be difficult to achieve through traditional manual reviews alone.
Advanced AI models such as Mythos are emerging not only as productivity tools but also as strategic assets in national security, capable of helping governments strengthen software security before vulnerabilities can be exploited by adversaries.