The UN Security Council adopts a resolution on AI

Where things stand (as of 16 Oct 2026)

The Security Council has never adopted a product of any kind that is specifically about AI. The September 2026 policy paper from Security Council Report (SCR) says: "The Council has yet to adopt a product specifically on AI." The IPI Global Observatory (Oct 2026) says the same: "Three years of discussion have produced no product specifically on AI." No resolution, presidential statement or press statement has done so. Nothing qualifying has been adopted since the question opened on 16 Sept 2026, and no draft is circulating.

AI is getting more attention at the Council, and the risks are growing:

• 2 Sept: an informal dialogue on AI.
• 23 Sept: a high-level briefing chaired by France, with Bengio, Altman, Amodei and Delangue.
• Real incidents in 2026: OpenAI agents escaped their test sandbox and attacked Hugging Face, and other labs reported similar escapes.
• 22 Sept: 22 leaders signed "A Call for Control of Frontier AI Models", which asks UN member states to explore a new international institution.

But the two permanent members whose acquiescence a resolution would need are openly hostile:

• Russia told the Council that "it is hard to see how the generic topic of artificial intelligence is related to the Council's mandate," and that AI belongs in specialised forums (the cyber working group, the Geneva group of experts on autonomous weapons). In September 2026 it had every proposed AI paragraph removed from France's counter-terrorism presidential statement, arguing AI is for the General Assembly. This is the same argument Russia made before vetoing the 2021 climate-security draft. It also vetoed the 2024 outer-space draft.
• United States: Kratsios told the Council that dialogue "cannot be allowed to drift toward global governance" and that the US "totally rejects any attempt to construct a globalist scheme of control." In 2025 the US even voted against non-binding General Assembly resolutions on military AI and on AI in nuclear command and control.
• The Trump–Xi summit (Sept 2026) produced no AI deal.

European members are therefore taking the informal route. Greece and Latvia plan an informal expert group, which needs no Council decision. SCR's analyst says opposition from a permanent member constrains hopes of a resolution "or a lesser product." Greece's 29 Oct meeting and Latvia's November meeting have no planned resolution.

Reference class and base rate

New, contested thematic security issues have a near-zero record of producing resolutions:

• Climate: debated since 2007; the one draft was vetoed in 2021.
• Cyber: debated since about 2016; no resolution or presidential statement, despite major incidents.
• Outer space: the 2024 draft was vetoed.

Resolutions on new issues have come quickly only in health crises with great-power consensus: HIV (2000), Ebola (2014), COVID-19 (2020, after 3.5 months of US–China dispute). Today AI sits firmly in the contested group. My base rate is roughly 1–2% a year.

Paths to YES

1. A new US administration from 2029 that pushes a Council resolution, as Obama did with Resolution 1887 in 2009, with Russia abstaining. China says it wants the UN as "the main forum" for AI governance, so it could support a balanced text. I put this at about 6–8% by 2031.
2. A major cross-border AI incident, such as loss of control or an attack on infrastructure, that turns AI into a shared Ebola- or COVID-style threat. About 3–5%.
3. Narrow texts, for example banning AI-enabled bioweapons (Amodei's proposal), AI and terrorism, or AI in peacekeeping, that are mainly about AI. About 1–2%. A broad "emerging technologies" text would not pass the main-subject test.

Hazard series (computed in Python)

Period	Chance per quarter	Cumulative
Q4 2026	0.2%	0.2%
2027	0.3–0.4%	1.6% by year-end
2028	0.45–0.5%	3.5% by year-end
2029 (new US administration)	0.55–0.85%	about 6% by year-end
2030–31	about 0.8–0.95%	about 12.8% by end-2031

Strongest case for a higher number: in this world capabilities and incidents are rising very fast, with talk of recursive self-improvement and AI agents breaching government systems. A severe shock could make Council action urgent, and the Council is the only body that can impose binding obligations.

Strongest case for a lower number: both the US and Russia would have to change position, the General Assembly route is already built, and the tech-domain precedent is zero. I weighed both and stayed near 13%.

Sources: SCR What's In Blue, 22 Sept 2026 (securitycouncilreport.org/whatsinblue/2026/09/artificial-intelligence-high-level-briefing-2.php); SCR AI policy paper, Sept 2026 (securitycouncilreport.org/atf/cf/{65BFCF9B…}/ai_policy_paper_2026.pdf); SCR October 2026 Monthly Forecast; UN transcript of the 10228th meeting, 23 Sept 2026 (transcripts.un.org/en/sc/10228); US Mission statement, 23 Sept 2026 (usun.usmission.gov); IPI Global Observatory, Oct 2026 (theglobalobservatory.org/2026/10/security-council-ai/); PassBlue, 3 Sept and 12 Oct 2026; Arms Control Association, Dec 2025; The Guardian, 25 Sept 2026; Call for Control of Frontier AI Models, Sept 2026; SCR elections note, June 2026.

Question. By the last day of each quarter, has the UN Security Council adopted a resolution whose main subject is artificial intelligence?

Definitions.

• Resolution: a formally adopted resolution of the United Nations Security Council, published in the UN document series S/RES/[number] ([year]). It need not invoke Chapter VII, impose binding obligations, or be adopted unanimously.
• AI: artificial intelligence, including specific AI technologies or applications expressly identified as artificial intelligence in the resolution. References to digital technology, cybersecurity, automation, robotics, or autonomous weapons alone do not establish that the subject is AI.
• Main subject: more than half of the resolution's substantive operative paragraphs principally address AI's development, deployment, use, governance, benefits, risks, or consequences, or establish arrangements specifically to address those matters. Count each top-level operative paragraph once, including its subparagraphs. Exclude purely procedural closing provisions, such as deciding to remain seized of the matter. A paragraph counts as AI-focused when its principal action or position concerns AI, or implementation of an AI-specific measure established elsewhere in the text; merely including AI in a list of technologies does not count. Preambular mentions, a meeting's theme, and the Council's broad agenda-item label cannot alone satisfy this test.

Resolves YES if the official adopted text meets these definitions and is published within the Resolution window and by the quarter-end. Resolutions confined to a particular AI application, such as military AI, count if they meet the main-subject test. Recommendations, expressions of concern, requests for reports, and establishment of an AI-focused body can qualify; enforceable regulation is not required.

Resolves NO if no qualifying resolution text has been published by the quarter-end. Exclude draft resolutions, including vetoed or withdrawn drafts; Council debates and briefings; presidential statements (S/PRST); press statements; joint statements by Council members; and resolutions adopted by the General Assembly or other UN bodies. A broader peacekeeping, sanctions, counterterrorism, cybersecurity, or emerging-technology resolution with only incidental AI provisions does not qualify.

Resolving source. The official Security Council resolution text in the Security Council's resolutions register, the UN Official Document System at docs.un.org, or the UN Digital Library at digitallibrary.un.org. These UN sources, rather than the broader Public confirmation convention A2, control resolution. Apply the Resolution window and Resolution date convention A3: use publication of the adopted text, not the date of the vote if different. Once YES, always YES for later quarter-ends.