6 KiB
| type | title | author | url | date | domain | secondary_domains | format | status | priority | tags | flagged_for_leo | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| source | REAIM Summit 2026 (A Coruña) — US and China Refuse to Sign, Only 35/85 Countries Endorse Military AI Principles | Multiple sources: TheDefenseWatch, US News, Asia Financial, Capacity Global | https://thedefensewatch.com/policy-strategy/us-and-china-refuse-to-sign-military-ai-declaration-at-reaim-summit/ | 2026-02-05 | ai-alignment |
|
news-coverage | unprocessed | high |
|
|
Content
The Third Summit on Responsible AI in the Military Domain (REAIM) was held February 4-5, 2026, in A Coruña, Spain.
Core finding: Only 35 out of 85 attending countries signed the commitment to 20 principles on military AI use ("Pathways for Action" declaration). The United States and China both declined to sign.
US position: The US signed the 2024 Seoul REAIM Blueprint for Action under Biden. Under Trump, at A Coruña 2026, Vice President J.D. Vance represented the US and declined to sign. Stated rationale: excessive regulation would stifle innovation and weaken national security. The shift represents a complete reversal of US multilateral military AI policy direction within 18 months.
China's position: China has consistently attended REAIM summits but avoided signing final declarations. Primary objection: disagreements over language mandating human intervention in nuclear command and control decisions. At A Coruña, China once again opted out.
Signatories: 35 nations including Canada, France, Germany, South Korea, United Kingdom, Ukraine. Notably: all middle powers, no AI superpowers.
Trend: Sharp decline from ~60 nations endorsing principles at Seoul 2024 to 35 at A Coruña 2026. The REAIM process, which was designed to build voluntary norms around military AI, is losing adherents, not gaining them.
GC REAIM Report: The Global Commission on Responsible AI in the Military Domain published its "Responsible by Design" report (September 24, 2025) seeking to translate REAIM Summit declarations into actionable guidance. The report presents three guiding principles and five core recommendations for all levels of the socio-technical AI lifecycle. Despite the quality of the report, the Third Summit saw dramatically reduced state participation.
Background on REAIM: Multi-stakeholder dialogue platform initiated by the Netherlands and South Korea, bringing together states, civil society, and industry to build shared norms for responsible military AI use. The platform was seen as a complementary track to the formal CCW GGE process.
Agent Notes
Why this matters: This is the clearest evidence of governance regression at the international level. The trend line is negative: 2022 (first REAIM, limited scope) → 2024 Seoul (60+ nations, US signs) → 2026 A Coruña (35 nations, US and China refuse). International voluntary governance of military AI is consolidating toward a smaller, less powerful coalition as the most advanced AI programs concentrate in non-participating states.
What surprised me: The magnitude of the decline. Going from 60 to 35 signatures in 18 months is a collapse, not a plateau. This is the international equivalent of Anthropic RSP rollback — voluntary commitment failure under competitive/political pressure, but at the international scale.
What I expected but didn't find: Any mechanism that could reverse the US position given the domestic political change. The Trump administration's rationale ("regulation stifles innovation") is precisely the alignment-tax race-to-the-bottom argument in diplomatic language. There's no near-term pathway to US re-engagement on multilateral military AI norms.
KB connections:
- the alignment tax creates a structural race to the bottom because safety training costs capability and rational competitors skip it — the US rationale for REAIM refusal is exactly this structural dynamic stated as policy
- voluntary safety pledges cannot survive competitive pressure — REAIM is the international case study for this mechanism: voluntary commitments erode as competitive dynamics intensify
- multipolar failure from competing aligned AI systems may pose greater existential risk than any single misaligned superintelligence — the competing US/China military AI programs represent the most dangerous multipolar scenario, and both are now outside any governance framework
- government designation of safety-conscious AI labs as supply chain risks — the same US government that blacklisted Anthropic for safety constraints is the one refusing REAIM principles
Extraction hints: Strong claim candidate: "International voluntary governance of military AI is experiencing declining adherence as the states most responsible for advanced autonomous weapons programs withdraw from multi-stakeholder norm-building processes — paralleling the domestic voluntary commitment failure pattern at the international level." This would extend the KB's voluntary commitment failure claim (currently documented domestically) to the international domain.
Curator Notes (structured handoff for extractor)
PRIMARY CONNECTION: voluntary safety pledges cannot survive competitive pressure because unilateral commitments are structurally punished when competitors advance without equivalent constraints WHY ARCHIVED: The REAIM 2026 outcome is the single clearest data point on international military AI governance regression. The trend (60→35 signatories, US reversal) documents the international layer of the voluntary commitment failure pattern. EXTRACTION HINT: Pair this with the UNGA 164:6 vote for the contrast: near-universal political expression (UNGA) coexists with sharp practical decline in voluntary commitments (REAIM). The gap between political expression and governance adherence is the key finding.