NSA Loses Access to Anthropic's AI Tool Amid Export Restrictions

The National Security Agency (NSA) has lost access to Anthropic’s advanced Mythos 5 AI model, which was instrumental in identifying software vulnerabilities. This development comes amid a protracted dispute between the US government and the Silicon Valley firm, as reported by the New York Times. The cutoff occurred after the Trump administration imposed export restrictions on Anthropic earlier this month, citing national security concerns.

The loss of access has deprived the NSA of a tool that had impressed its analysts with its ability to detect software weaknesses rapidly. During testing, the Mythos 5 model reportedly identified vulnerabilities in highly secure government networks within hours, according to an anonymous US official cited by the Associated Press.

Anthropic's AI technology has been increasingly utilized within classified government networks, playing a role in intelligence analysis, operational planning, and cyber operations. However, the Department of War designated Anthropic a “supply chain risk” in February after the company refused to lift restrictions on certain military applications of its AI systems. Anthropic has publicly opposed mass domestic surveillance and fully autonomous weapons, leading to a federal directive from President Trump for agencies to phase out the use of Anthropic technology within six months.

In response, Anthropic filed a lawsuit against the government, claiming that the restrictions were unlawful retaliation for its refusal to relax safeguards on military uses of AI. Despite the phase-out order and the ongoing legal battle, reports have indicated that some parts of the US government continue to use Anthropic’s systems.

These developments occur amid growing concerns from researchers, technology leaders, and security officials about the rapid integration of AI systems into military and intelligence operations. Experts warn that the same AI tools designed to enhance cyber defenses could also facilitate automated attacks, lowering the barriers for malicious actors. Recently, cybersecurity experts from the Five Eyes intelligence alliance, which includes the US, UK, Australia, Canada, and New Zealand, cautioned that advancing AI models could soon empower hackers to disrupt governments, businesses, and critical infrastructure on a global scale.