This week’s defence AI news is defined by two converging pressures: governments racing to embed AI deeper into classified military infrastructure while simultaneously grappling with who sets the rules — vendors, contracts, or law. Meanwhile, the UK is edging toward removing humans from lethal weapons decisions, and Europe’s counter-terrorism coordinator is sounding the alarm over AI-generated extremist content outpacing enforcement.
Top story: OpenAI has struck a deal with the Pentagon to deploy its models on classified networks with safety guardrails intact — setting a new template for how AI labs can engage with military clients without stripping ethical safeguards.
OpenAI Cuts Pentagon Deal, Keeps Safety Stack Intact
OpenAI (official blog) · Strategy
OpenAI has reached an agreement with the US Department of Defense to deploy advanced AI in classified environments, but unlike competitors it is retaining full control of its safety stack and keeping cleared personnel in the loop. The deal explicitly prohibits domestic surveillance of US persons and sets contractual red lines the DoD cannot cross. It matters because OpenAI is positioning its layered safeguard approach as the responsible industry template — and has asked that the same terms be made available to all AI companies working with the Pentagon.
https://openai.com/index/our-agreement-with-the-department-of-war/
Pentagon Ditches Single-Vendor AI, Goes Multi-Cloud for War
Federal News Network · Risk
The Pentagon is overhauling its AI procurement model, shifting from reliance on a single provider toward a modular, multi-vendor cloud architecture spanning OpenAI, Google, Microsoft, Nvidia, SpaceX, AWS, Oracle, and Reflection. The move follows its public fallout with Anthropic and is designed to eliminate vendor lock-in on classified networks. For defence practitioners, the shift signals that open-weight, inspectable models are becoming a security requirement — not just a cost option — because closed systems the military cannot audit create accountability gaps in rules-of-engagement decisions.
UK Minister Floats Removing Humans From Lethal Weapons Loop
RT (citing UK MoD statements) · Regulation
UK Armed Forces Under-Secretary Al Carns has publicly argued that Britain may need to reconsider its policy requiring human involvement in weapons that identify, select, and attack targets — suggesting adversaries would not hesitate to deploy fully autonomous lethal systems. The UK MoD had already launched a legal review of its autonomous weapons framework in February 2026, describing existing rules as no longer ‘fit for the current era of threat.’ The statement is one of the most explicit signals yet from a Western government that the human-in-the-loop principle is under active revision, with profound implications for international humanitarian law.
EU Counter-Terrorism Chief: AI Propaganda Is the New Front Line
EU Perspectives · Risk
EU Counter-Terrorism Coordinator Bartjan Wegter warned the European Parliament that AI is enabling terrorist organisations and lone-actor radicals to produce sophisticated propaganda ‘with unprecedented ease,’ describing AI-generated manifestos and recruitment materials as the new front line of radicalisation. He stressed that Europe’s core problem is no longer a lack of legislation — the Digital Services Act, Terrorist Content Online Regulation, and AI Act all exist — but an implementation gap as technology moves faster than enforcement. The warning is a significant escalation in EU-level concern and points to growing pressure on platforms and member states to operationalise existing rules before the next wave of AI-enabled extremism.
https://euperspectives.eu/2025/12/ai-extremism-counter-terrorism-coordinator/
Lawfare: Military AI Governance by Contract Is Dangerously Inadequate
Lawfare · Legal
A detailed Lawfare analysis argues that the US has drifted into ‘regulation by contract’ for military AI — where the rules governing battlefield AI are set not by statute or democratic deliberation, but by bilateral agreements between the DoD and individual vendors that bind only the parties who signed them. The piece examines how the Anthropic-Pentagon standoff exposed a structural flaw: when the Pentagon mandates ‘any lawful use’ language and frames governance requirements as deployment blockers, oversight becomes something that must be negotiated rather than guaranteed. For UK and European defence practitioners, the piece is a timely warning as allied nations consider their own procurement frameworks.