President Donald Trump signed National Security Presidential Memorandum 11 on Friday, directing the US military and intelligence agencies to accelerate their adoption of advanced AI while protecting frontier models from theft by foreign adversaries. The directive replaces the Biden administration’s NSM-25, which had governed AI in national security since 2024, and adds a provision that no commercial vendor can disable, degrade, or modify an AI system that American warfighters depend on without prior government approval.

The memo lands in a week already crowded with AI policy moves: a voluntary executive order on frontier model reviews, the announcement that Anthropic is preparing to release Mythos-class models to the general public, and growing pressure from Senator Elizabeth Warren to probe whether the administration’s approach is moving too fast without adequate safeguards.

What NSPM-11 requires

The memo gives agencies 120 days to review and update procurement processes so they can onboard advanced AI models from multiple vendors rapidly. It requires the Secretary of Defense to update the directive governing autonomy in weapon systems within 90 days. It calls on agencies including the FBI, the Office of the Director of National Intelligence, and the Office of the National Cyber Director to maintain “deep, proactive” partnerships with the AI industry.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

The language is blunt: US military and intelligence agencies have been too slow to adopt AI, and the technology is developing faster than their procurement systems can accommodate. The memo also discloses the existence of a classified annex, to be issued within 90 days, whose contents are not public.

On the defensive side, the memo directs officials including Defense Secretary Pete Hegseth and NSA chief General Joshua Rudd to work with “willing private-sector companies” to develop protocols protecting frontier models from adversary efforts to steal or replicate them. The distillation attacks the White House described in April, where Chinese labs submitted millions of queries to American models to replicate their capabilities, are the explicit concern.

The Mythos factor

NSPM-11 arrives as Anthropic prepares to release Mythos-class models to the general public in the coming weeks. Mythos found more than 10,000 high-severity software vulnerabilities in its first month of testing and was initially shared only with a limited group of trusted partners through Project Glasswing. Making it widely available means the same offensive cybersecurity capabilities that the US government wants to deploy will also be accessible to anyone with an API key.

CISA, the top US cyber defence agency, is expected to issue its own directive in the coming days ordering federal agencies to secure their networks against AI-boosted hackers. The layering of NSPM-11, the voluntary executive order, and the coming CISA directive represents the most concentrated burst of AI security policy the administration has produced.

Warren pushes back

Senator Elizabeth Warren sent a letter to Senate Banking Committee Chair Tim Scott on Monday, calling for hearings with Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent on the administration’s AI policy. Warren argued that the voluntary model-review provision in the executive order gives AI companies a “free pass at the expense of Americans’ safety and security.”

She also pressed on chip exports. Warren cited Commerce Department guidance issued last month to block Chinese companies from acquiring advanced AI chips through overseas affiliates, calling the existing export control regime riddled with loopholes. “The Commerce Department’s mismanagement of US export controls has resulted in loopholes that may have allowed the most advanced AI chips to flow to companies with ties to China’s military,” she wrote.

The two-speed problem

The administration is simultaneously telling agencies to move faster on AI adoption and trying to protect the models those agencies will use from the adversaries who want to copy them. It is building a framework that is voluntary for the companies, mandatory for the agencies, and contested by Congress.

Meanwhile, Iranian drone strikes earlier this year knocked out power to Amazon Web Services data centres in Bahrain and the UAE, demonstrating that the physical infrastructure AI runs on is itself a target. The memo calls on agencies to work with the private sector to boost both physical and cybersecurity of data centres.

Sriram Krishnan, one of the administration’s top AI advisers, announced on Friday that he will leave the White House at the end of the month. His departure removes one of the few figures who bridged the gap between Silicon Valley and the national security establishment at a moment when that bridge has never been more load-bearing.