This week’s defence AI stories reveal a world moving faster than its governance frameworks: the US has formally codified military AI acceleration via NSPM-11 while the EU rushes to finalise its AI Act omnibus amendments ahead of August deadlines. The real-world stress test of Palantir’s Maven system in Operation Epic Fury continues to generate accountability debates, underscoring that doctrine and ethics are lagging dangerously behind deployment.
Top story: Trump signed NSPM-11 on 5 June, the most sweeping US national security AI directive ever, ordering accelerated military AI adoption, a 90-day rewrite of autonomous weapons policy, and contract termination powers over non-compliant AI vendors.
Trump’s NSPM-11 Rewrites US Military AI Doctrine in One Memo
Council on Foreign Relations · Strategy
On 5 June, President Trump signed National Security Presidential Memorandum 11, directing every military and intelligence agency to accelerate AI adoption, ordering rapid onboarding of advanced models from multiple vendors, and mandating a 90-day rewrite of DoD Directive 3000.09 — the Pentagon’s foundational policy on autonomous weapons. The memo also requires agencies to terminate contracts with AI companies that repeatedly limit government use of their technology, a direct response to the Anthropic standoff, and bars any commercial entity from disabling or modifying AI systems warfighters depend on without prior approval. For defence practitioners, NSPM-11 signals a decisive shift: AI governance accountability now flows through the military chain of command rather than external regulators or commercial ethics constraints.
https://www.cfr.org/articles/what-trumps-national-security-ai-memo-gets-right-and-leaves-unresolved
Maven’s Iran War Debut Exposes AI Targeting’s Accountability Gap
Arms Control Association · Risk
Detailed post-operation disclosures confirm that Palantir’s Maven Smart System was used to help plan and coordinate over 13,000 airstrikes during Operation Epic Fury, processing data from radar, satellite imagery, and drone feeds to generate strike options and courses of action at unprecedented speed. Usage metrics were staggering — classified usage surged 89% month-on-month, with peak daily token consumption hitting 20 billion — but reports of civilian casualties, including a school strike, have intensified calls for stricter human oversight of AI-assisted targeting. The episode is being widely cited as the first large-scale combat deployment of AI decision-support at this scale, raising urgent questions about whether ‘human-in-the-loop’ assurances remain meaningful when the pace of AI recommendation far outstrips human review capacity.
https://www.armscontrol.org/act/2026-05/news/ai-plays-major-role-war-iran
EU Strikes AI Act Omnibus Deal, Pushes High-Risk Deadlines Back
Global Policy Watch · Regulation
EU negotiators reached a provisional agreement on 7 May on the Digital Omnibus on AI — the first set of amendments to the EU AI Act since its 2024 adoption — deferring compliance deadlines for high-risk AI systems by up to 16 months, with obligations for use-based systems now pushed from August 2026 to December 2027. The deal is directly relevant to defence-adjacent sectors: AI used in biometrics, critical infrastructure, and border control falls within the high-risk categories affected by the deferral. With formal adoption expected in June and publication in July, defence and security organisations operating in Europe now have a clearer — if extended — runway to prepare compliance programmes before the next major milestone.
Anthropic’s Mythos AI Cleared for NSA But Legal Battle Continues
Defense One · Legal
Joint Chiefs Chairman Gen. Dan Caine declared autonomous weapons will be a ‘key and essential part of everything we do,’ even as the legal fallout from the Pentagon-Anthropic dispute remains live: a federal judge issued a temporary injunction against Anthropic’s ‘supply-chain risk’ designation after the company refused to drop contractual limits barring Claude from fully autonomous weapons and domestic mass surveillance. Meanwhile, Anthropic’s powerful new frontier model, Mythos Preview, has been granted access to the NSA, signalling that even as the broader contract dispute drags through the courts, intelligence community demand for cutting-edge AI is overriding procurement friction. The standoff is reshaping how AI vendors across the industry think about government use-case clauses.
https://www.defenseone.com/policy/2026/04/autonomous-weapons-warfare-joint-chiefs/413065/
UK Government Commits £90m and Calls AI Firms to Cyber Defence
Cyber Magazine · Strategy
The UK Security Minister issued a public call to action for leading AI companies to collaborate with the government on building national AI-powered cyber defence capabilities, backed by a £90 million commitment to bolster the defences of small and medium-sized businesses. The move comes as NCSC CEO Richard Horne warned at CYBERUK 2026 that the number of security incidents managed by the centre more than doubled in 2025, driven by hostile states deploying AI tooling against UK infrastructure. The Cyber Resilience Pledge — already sent to over 180 major business leaders — reflects a growing UK posture that national cyber defence is inseparable from frontier AI capability, putting the country’s approach in direct dialogue with NATO allies scaling similar public-private defence AI partnerships.
https://cybermagazine.com/news/what-is-uk-governments-plans-for-national-cyber-defence