White House issues NSPM-11 on AI in national security

President Donald Trump signed National Security Presidential Memorandum 11, directing military, intelligence and other national-security agencies to accelerate adoption of AI across warfighting and intelligence work. The memorandum is organised around four stated pillars: adoption (removing barriers to rapid deployment), adaptation (drawing on commercial and open-source AI rather than only bespoke government systems), assurance (that systems deployed are reliable and controllable) and accountability (guarding against censorship, ideological bias and unlawful surveillance in the systems used).

Its most consequential procurement provision instructs agency heads to terminate, for default or convenience, contracts with AI companies — and their subcontractors — found to have “repeatedly demonstrated a pattern of conduct” inconsistent with the memorandum’s policies, with only narrow, time-limited waivers available. Read against Dario Amodei’s public refusal months earlier to give the Pentagon unfettered use of Claude — a dispute that preceded OpenAI signing its own agreement with the Department of War — the memorandum reads as an attempt to foreclose that kind of vendor-imposed restriction across the industry, not just at Anthropic. It formally rescinds and replaces NSM-25, the Biden administration’s 2024 national-security AI memorandum, which had paired adoption with more extensive governance guardrails.

Agencies have 90 to 120 days to update relevant directives and procurement processes, including the Pentagon’s policy on autonomy in weapons systems. Supporters characterised the memorandum as removing bureaucratic drag on national-security AI adoption; critics, including the Electronic Privacy Information Center, argued it stripped away guardrails the prior administration had put in place without replacing them with comparably specific safeguards.