On February 4-5, the third Responsible AI in the Military Domain (REAIM) summit convened at the PALEXCO center in A Coruña, Spain. Reuters reported the outcome on February 5: of the 20-point joint declaration on the table, only about a third of the 85 attending countries — 35 nations — signed. The United States and China, the two AI superpowers, both opted out.
The significance fits in one sentence: the pace of military AI adoption has now clearly outrun the pace of global cooperation on governing it. When both great powers sit outside an accountability framework, the declaration binds none of the countries actually developing advanced military AI — and it was never legally binding to begin with. Its entire force was political weight, which is exactly what walked out.
What the 20-Point Declaration Says
Per Reuters and Capacity, the declaration’s core principles include:
- Maintaining human oversight of AI-enabled weapons
- Establishing clear chains of command
- Robust risk assessments for military AI systems
- Sharing national-level oversight mechanisms
- Training for personnel who use military AI
Signatories include Canada, Germany, France, Britain, the Netherlands, South Korea, and Ukraine, among the 35. Notably, the US and China did not skip the summit — both sent senior delegations that participated in the discussions, then declined to sign. Per Capacity’s reporting, the refusals came despite heavy diplomatic participation from both countries across the two days, which makes the opt-out a considered position rather than an absence.
Why Washington and Beijing Stayed Out
Delegates attributed the refusals to the realities of strategic competition. One factor is uncertainty about how transatlantic alliances will evolve, which makes multilateral commitments less attractive. The other is fear that signing would weaken national competitive advantage — when your rival is not bound by the same rules, self-restraint amounts to unilateral disarmament. Both reasons apply to the US and China, and to a number of the non-signatories as well.
Dutch defence minister Ruben Brekelmans put the structure most plainly in his comments to Reuters. He described the situation as a “prisoner’s dilemma”: governments want responsible AI use, but fear falling behind as rivals accelerate. “Russia and China are moving very fast. That creates urgency to make progress in developing AI,” he said — and the same speed raises the urgency of working on responsible use, with the two going “hand-in-hand.”
From The Hague to A Coruña: A Shrinking Consensus
Zoom out and this partial signature is the newest point on a downward line. At the first REAIM summit in The Hague in 2023, both China and Russia endorsed the “call to action” document; this year in A Coruña, China opted out entirely, withdrawing even that symbolic support. An analysis by the Council on Foreign Relations argues that with US and China engagement fading, middle powers face a binary: lead the conversation themselves, or wait — while actual military AI deployment waits for no one. It is worth remembering what REAIM has always been: a non-binding exercise whose product is political language, not law. That was workable when the signatories included the states building the most capable systems. Without them, the declaration sets norms for the countries least likely to determine how military AI is actually used.
What It Means for the AI Industry
This is not just diplomatic news. First, defense and dual-use AI vendors face a more fragmented rulebook: as 35 signatories each translate the declaration’s principles into national policy, contractors will confront 35 overlapping but unequal accountability regimes, plus two major markets entirely outside the framework. Second, the declaration’s emphasis on human oversight, command chains, and risk assessments will bleed into commercial contracts through procurement requirements and due diligence even without legal force — AI teams selling to governments should read the 20 points as a de facto standard, and a reasonable preview of what NATO-aligned buyers will ask about in 2026 tenders. Third, the prisoner’s dilemma has no near-term exit: without verification mechanisms, no side can afford to slow down unilaterally. That matches the governance-fragmentation trend we flagged in our 2026 opening outlook — the rules are splitting along national borders, not converging along technical standards.
Sources
- US, China opt out of joint declaration on AI use in military — Reuters
- US and China snub military AI agreement at global summit — Capacity
- Military AI Adoption Is Outpacing Global Cooperation — CFR
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.