Despite the U.S. Department of Defense labeling artificial intelligence (AI) firm Anthropic as a supply chain risk, reports suggest that the National Security Agency (NSA) and certain security organizations are employing the company’s latest AI model, Claude Mythos Preview. This reveals a discrepancy between the Pentagon’s official stance and actual security practices in the field.
According to reports from Axios and other media outlets on Tuesday, the NSA is internally using Claude Mythos Preview, Anthropic’s most advanced AI model. This usage contradicts the Defense Department’s efforts to restrict Anthropic’s involvement due to supply chain concerns.
In February, the Pentagon moved to terminate collaboration with Anthropic and urged related companies to follow suit. This matter has since escalated into a legal dispute.
While the Defense Department aims to limit use, the NSA continues to leverage the technology. Axios reports that as cybersecurity demands grow within military and intelligence circles, Anthropic’s tech is seeing increased adoption.
Multiple sources cited by Axios indicate that beyond the NSA, several defense-related entities are increasingly implementing this model.
Claude Mythos is primarily used to identify security vulnerabilities in corporate and government systems. Its exceptional ability to detect vulnerabilities and assess potential exploits has led to restricted access, with only select institutions granted use rather than public availability.
Axios reported that Anthropic has limited Claude Mythos’s powerful cyber capabilities to roughly 40 institutions, including the NSA. British intelligence agencies are also said to access the model through their AI Security Institute.
This situation highlights a divergence between the Pentagon’s regulatory stance and actual security operations. While the government warns of risks, there’s a growing trend of using Claude Mythos as a security tool in the field.
Limited Release of Claude Mythos Gains Traction in Security Operations
Dialogue between Anthropic and the White House continues. Axios reported that Anthropic Chief Executive Officer (CEO) Dario Amodei recently met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent to discuss AI use in government and related security issues. Both sides reportedly found the meeting productive, with discussions including potential applications beyond the Defense Department.
Concerns about Claude Mythos are spreading to state governments. Axios reported that Maryland Governor Wes Moore planned a private dinner on Monday with tech industry representatives, including Microsoft, to discuss cybersecurity strategies in the Claude Mythos era.
Maryland hosts Fort Meade, home to NSA and U.S. Cyber Command headquarters. This discussion shows AI security issues are becoming local policy concerns, extending beyond federal debates.
The rift between Anthropic and the U.S. government stems from differing views on AI technology use. The Pentagon previously demanded Anthropic open its AI model for all legal uses, including military applications. However, the company maintains that it should restrict uses for mass surveillance or autonomous weapons development.
South Korea Conducts Urgent Reviews as Government and Industry Responses Intensify
In South Korea, efforts to assess security threats and response systems related to advanced AI proliferation are gaining momentum.
On April 14, the Financial Supervisory Service convened an emergency meeting with cybersecurity officials from major financial institutions to discuss countermeasures. The National Security Office reportedly ordered urgent responses from relevant public and private agencies. The Ministry of Science and Information and Communications Technology (ICT) held four emergency meetings over April 14-15.
On April 14, the Ministry of Science and ICT held an emergency meeting with Chief Information Security Officers (CISOs) from major telecom companies, NAVER, Kakao, Woowa Brothers, and Coupang to address pressing issues. The same day, the Ministry also met with domestic AI security experts.
On April 15, the Ministry gathered major domestic security firms in the morning, followed by an afternoon meeting with CISOs from 40 leading companies, including SK hynix, Shinhan Bank, Hyundai Card, Samsung Medical Center, Kakao Healthcare, and Lotte Shopping.
Meeting participants agreed that AI-based attacks could materialize faster than expected, emphasizing the need to bolster both security threat responses and AI-based defense capabilities.
The Ministry of Science and ICT stated it continues to monitor international trends and domestic company situations while maintaining ongoing communication with the industry.
