Summary: At the third REAIM summit in A Coruña on February 4-5, 2026, only 35 of 85 attending countries signed the non-binding 20-point declaration on responsible military AI. The United States and China sent senior delegations and participated but did not sign, leaving the countries building advanced military AI outside the political framework. That outcome underscores the article’s warning that adoption is outpacing governance, with no near-term exit from the competitive prisoner’s dilemma.
The third Responsible AI in the Military Domain (REAIM) summit ran February 4-5 at the PALEXCO center in A Coruña, Spain. Reuters carried the outcome on February 5: of the 85 countries attending, only about a third — 35 nations — signed on to the 20-point joint declaration that had been put before them. The United States and China, the two AI superpowers, left without either adding a signature.
Strip the story to one line and here it is: adoption of military AI is now decisively ahead of the world’s attempts to cooperate on governing it. With both great powers on the outside of the accountability framework, nothing in the declaration reaches the countries actually building advanced military AI — and the document was never legally binding to begin with. Political weight was its only engine, and political weight is exactly what walked out.
The Declaration, Point by Point
Assembled from the Reuters and Capacity coverage, the declaration’s core principles read:
- Human oversight kept in place for AI-enabled weapons
- Command chains defined without ambiguity
- Military AI systems put through rigorous risk assessments
- National-level oversight mechanisms shared across states
- Training for the personnel running military AI
The 35 signatories count Canada, Germany, France, Britain, the Netherlands, South Korea, and Ukraine among their number. What bears noting: the US and China did not skip the event — both sent senior delegations into the discussions, and both stopped short of signing all the same. Per Capacity, the two countries were diplomatically engaged throughout the two days before they declined, casting the refusal as a deliberate position rather than an empty chair.
Why the Two Powers Held Back
Delegates talking to the press located the refusals in strategic competition as it operates in practice. One factor is not knowing where the transatlantic alliances are headed, which takes the gloss off multilateral commitments. The other is the fear that a signature would cost a country its competitive edge — when your rival answers to none of the same rules, restraint is unilateral disarmament under another name. The US and China match both descriptions, as do several other countries that declined to sign.
Dutch defence minister Ruben Brekelmans, speaking to Reuters, exposed the structure as sharply as anyone. He reaches for a “prisoner’s dilemma”: responsible AI use is what every government says it wants, yet none will risk falling behind while rivals accelerate. “Russia and China are moving very fast. That creates urgency to make progress in developing AI,” he observed — the same speed that lends development its urgency makes responsible use just as time-sensitive, with the two traveling “hand-in-hand.”
From The Hague to A Coruña: A Consensus in Retreat
Widen the timeframe and this partial signature is simply the newest marker on a downward slope. At the first REAIM summit in The Hague in 2023, the “call to action” document still won the backing of China and Russia both; by this year’s gathering in A Coruña, China had withdrawn entirely, taking even that gesture of support off the table. The Council on Foreign Relations argues that as US and Chinese engagement thins out, middle powers confront a binary: keep the conversation alive on their own, or wait — while the deployment of real military AI waits for nobody. It helps to remember what REAIM has always been: a non-binding exercise that yields political language instead of law. That setup functioned as long as the signatories counted among them the states building the most capable systems. With those states gone, the declaration is drafting norms for the countries with the least influence over how military AI actually gets used.
What This Means for AI Companies
The stakes are not confined to diplomacy. For defense and dual-use AI vendors, the first consequence is a splintering rulebook: as each of the 35 signatories converts the declaration’s principles into national policy, contractors will be managing 35 accountability regimes that overlap without aligning — and that is before counting two major markets that sit entirely outside the framework. The second consequence involves the declaration’s emphases — human oversight, command chains, and risk assessments — which, with or without legal force, will find their way into commercial contracts through procurement requirements and due diligence; AI teams selling to governments would be wise to read the 20 points as the working standard, and as a reasonable preview of what NATO-aligned buyers will scrutinize in 2026 tenders. Third, the prisoner’s dilemma has no near-term exit: without verification mechanisms, no side can risk slowing down alone. All of which matches the trend toward fragmented governance we pointed to in our 2026 opening outlook — rules are cracking along national lines instead of converging around technical standards.