The US National Security Agency (NSA) has lost access to Anthropic’s advanced Mythos 5 AI model while using it to find software weaknesses, the New York Times has reported. The development comes amid Washington’s months-long dispute with the Silicon Valley firm.

The cutoff came after the Trump administration imposed export restrictions on Anthropic earlier this month, citing national security concerns, according to the NYT.

The loss “deprived” the intelligence agency of a “tool that has impressed and alarmed its analysts with how good it is at finding software weaknesses,” the outlet added.

During the tests, the model identified vulnerabilities in highly secure government networks “within hours,” the AP reported on Wednesday, citing an anonymous US official.

Anthropic’s AI technology has been increasingly deployed on classified government networks and integrated into US national-security work, with its models used for intelligence analysis, operational planning, and cyber operations.

However, in February, the Department of War designated Anthropic a “supply chain risk” after the company refused to remove restrictions on some of its AI systems’ military applications. The firm said it opposed mass domestic surveillance and fully autonomous weapons. US President Donald Trump subsequently ordered federal agencies to phase out Anthropic technology within a six-month window.

Anthropic sued the government, arguing that the measures were unlawful retaliation for its refusal to relax safeguards on AI’s military uses.

Despite the phase-out order and ongoing legal battle, multiple media reports later claimed that parts of the US government continue to use Anthropic systems.

The developments come amid warnings from researchers, technology leaders, and security officials that AI systems are being integrated into military and intelligence operations faster than governments and institutions can adapt to their increasing capabilities. Experts have cautioned that the same tools that are used to strengthen cyber defenses could also automate attacks and lower barriers for malicious actors.

On Monday, cybersecurity experts from the Five Eyes intelligence alliance – the US, UK, Australia, Canada, and New Zealand – warned that rapidly advancing AI models could soon enable hackers to disrupt governments, businesses, and critical infrastructure on a global scale.