Anthropic embeds engineers at US security agency to deploy Mythos AI: Report

Anthropic, a San Francisco-based AI company, has embedded engineers within the U.S. National Security Agency (NSA) to integrate and adapt its advanced Mythos AI model for cybersecurity and national security applications. The deployment, reported by the Financial Times, involves a small team of 'forward-deployed engineers' working directly with the NSA to tailor Mythos for specialized operational needs, including potential use in penetrating foreign networks linked to rivals like China and Iran. The collaboration occurs amid legal tensions between Anthropic and the Pentagon, which classified the company as a 'supply-chain risk' due to disputes over restrictions on its Claude AI models. Anthropic had sought to bar their use in mass domestic surveillance and autonomous lethal weapons, but the Pentagon’s designation could limit its military contracts. Despite the legal battle, Anthropic has continued cooperating with national security agencies, particularly after launching Mythos—a cybersecurity-focused AI designed to identify software vulnerabilities. Mythos was first introduced earlier this year and has since gained attention from governments, banks, and tech firms for its ability to detect and analyze security flaws. Critics warn that its capabilities could also be used offensively, enhancing cyber operations. Earlier this week, Anthropic expanded Mythos access to approximately 150 organizations across more than 15 countries, broadening its reach beyond early adopters in the U.S. and U.K. The move reflects intensifying competition in AI-driven cybersecurity, with rivals like OpenAI developing similar tools. While supporters argue Mythos strengthens digital defenses, its dual-use potential raises ethical concerns. Neither Anthropic nor the U.S. Department of Defense has commented on the NSA deployment, leaving the full scope of the collaboration unclear.