The National Security Agency has quietly started using Anthropic's latest AI model, Mythos Preview, even as some Pentagon officials remain wary of the company's supply chain risks. It's a move that highlights the government's increasingly tricky balancing act between security concerns and the race for technological advantage.
According to sources familiar with the matter, the NSA is using Mythos Preview to scan its internal systems for potential cybersecurity vulnerabilities—a strictly defensive application. Anthropic had previously kept tight controls on the model, granting access to only about 40 organizations. The reason? Its powerful capabilities related to cyber offense make it a sensitive tool.
This adoption spotlights a growing rift within the U.S. government over how to approach AI. On one side, some defense officials are skeptical of Anthropic's stance on military applications, particularly its restrictions around mass surveillance and autonomous weapons. On the other, there's a push to secure cutting-edge AI capabilities now to bolster cyber defense and intelligence analysis.
Anthropic is now considered one of the top-tier AI developers alongside OpenAI and Alphabet's Google. If its models gain traction within U.S. intelligence agencies, it could significantly boost the company's standing in the government market and accelerate the commercialization of its tech in the public sector.
Meanwhile, Anthropic's leadership has recently met with senior White House officials to discuss how the government uses its AI systems and the safety controls in place. Those talks were described as "constructive," and they could pave the way for broader adoption across more agencies.