The Security Council adopts a binding Chapter VII resolution on AI

Where things stand (12 Oct 2026)

The Council has never adopted a product of any kind specifically on AI, binding or not. Security Council Report's September 2026 policy paper says plainly: "The Council has yet to adopt a product specifically on AI." The only technology-specific product is a 2024 presidential statement (S/PRST/2024/6). Even that lost its operational language before adoption. IPI's 7 Oct 2026 analysis says the same: three years of meetings have produced no AI product.

AI language is still being stripped even from non-binding texts. In September 2026, Russia had a paragraph on terrorists' misuse of AI deleted from France's counter-terrorism presidential statement, arguing that AI belongs in the General Assembly. The final statement did keep a general line: the Council intends to address "misuse of technologies" in future counter-terrorism resolutions.

All three key permanent members currently oppose Council rules on AI. At the 23 Sept 2026 briefing (S/PV.10228, with Altman, Amodei and Bengio):

• United States (Kratsios): it "totally rejects any attempt to construct a globalist scheme of control of superintelligence"; national legislatures should regulate and "you should do the same."
• Trump, separately, called AI-risk concerns a "hoax" and a "sick conspiracy."
• Russia: "hard to see how the generic topic of AI is related to the Council's mandate"; such issues belong in specialised forums.
• China: favours "consensus-based" governance led by the UN as a whole, criticises "cliques" and "digital hegemony," and promotes its own AI organisation, WAICO.

The Trump–Xi summit (24–26 Sept) produced only an AI dialogue channel.

The Council is unusually divided and unproductive. Russia and China vetoed the US text on the Iran sanctions panel (18 Sept) and blocked expanding the Darfur arms embargo (Oct). I downloaded S/RES/2815–2829 (2026) from the documents.un.org symbol API. Only 15 resolutions were adopted from 30 Jan to 29 Sept 2026, and none mentions artificial intelligence. Resolution 2823 refers only to "emerging technologies," and it is not a Chapter VII resolution.

Reference class

New Chapter VII "legislative" obligations covering whole topics are rare: 1373 on terrorism (2001), 1540 on weapons of mass destruction and non-state actors (2004), 2178 on foreign terrorist fighters (2014), 2396 (2017), 2462 (2019), and the 2664 humanitarian exemption (2022). All needed agreement among the permanent members, usually under US leadership, often after a shock.

Cyber is the closest technology comparison: about 25 years of salience and no resolution at all. Other thematic crises such as Ebola and COVID produced only non–Chapter VII, non-binding resolutions. My outside view is roughly 1% a year for a fast-rising technology domain, lower while the US, Russia and China all object.

Paths to YES

1. Counter-terrorism and sanctions texts.
   • Resolution 2734 (2024), adopted under Chapter VII, says asset-freeze duties cover resources used for "Internet hosting and related services," but it never mentions AI. Its Monitoring Team and Ombudsperson mandates expire in June 2027, and the US holds the pen. Adding AI to a binding clause there is the most concrete near-term route, but it is unlikely given Russia's objection and the US preference for streamlined renewals. I put about 0.5% on that quarter.
   • Resolution 2396 (2017) shows that binding technology-specific duties (biometrics) can appear in counter-terrorism texts.
   • Under resolution 2663, the first comprehensive review of resolution 1540 is due before December 2027. The 2016 review led to Chapter VII resolution 2325, but its technology language was only "calls upon."
2. A shock-driven or deal-driven thematic resolution. Examples: an AI-enabled terrorist or bioweapon attack, a catastrophic incident involving autonomous AI agents (2026 already saw sandbox escapes and an agent attack on Hugging Face), or a US–China AI deal that both want to make binding on everyone, as resolution 1540 did. This becomes likelier after January 2029 if US policy changes.
3. Country-specific sanctions that name AI (for example, AI chips). This is improbable because Russia and China veto new sanctions.

Paths to NO

The Trump administration is hostile until January 2029. Russia objects to the Council's competence over AI in principle. China wants General Assembly forums instead. Council output is low. The bar is high: the binding provision itself must expressly cover AI.

Forecast

The quarterly hazard is 0.07–0.17% in most quarters to 2028. It rises to about 0.25–0.36% in 2029–31 to reflect a possible change of US administration and growing salience. Bumps of 0.5% fall in Q2 2027 and Q2 2030 for the ISIL/Al-Qaida sanctions renewals, assuming 36-month cycles. I computed the cumulative series as 1 − ∏(1 − h): about 0.9% by end-2027, 1.5% by end-2028, and 5.3% by end-2031.

Strongest case against my estimate: AI is moving faster than cyber ever did, and a single catastrophe could produce a resolution like 1373 within weeks. I allow for this in the later quarters but do not go higher, because even a shock would have to overcome rivalry between the US and China and Russia's objection that AI is outside the Council's remit.

Sources

• IPI Global Observatory, 7 Oct 2026: https://theglobalobservatory.org/2026/10/security-council-ai/
• Security Council Report, What's in Blue, 22 Sept 2026: https://www.securitycouncilreport.org/whatsinblue/2026/09/artificial-intelligence-high-level-briefing-2.php
• Security Council Report AI policy paper, 15 Sept 2026: https://www.securitycouncilreport.org/atf/cf/%7B65BFCF9B-6D27-4E9C-8CD3-CF6E4FF96FF9%7D/ai_policy_paper_2026.pdf
• Security Council Report on the counter-terrorism presidential statement, 11 Sept 2026: https://www.securitycouncilreport.org/whatsinblue/2026/09/counter-terrorism-briefing-to-commemorate-the-25th-anniversary-of-the-9-11-attacks.php
• S/PV.10228 transcript, 23 Sept 2026: https://transcripts.un.org/en/sc/10228
• Resolution texts via https://documents.un.org/api/symbol/access?s=S/RES/2734(2024)&l=en&t=pdf (also 2663, 2325, 2396, 2810, and 2815–2829 (2026))
• 1267 Monitoring Team mandate: https://main.un.org/securitycouncil/en/sanctions/1267/monitoring-team/work-and-mandate
• Security Council Report on resolution 2810: https://www.securitycouncilreport.org/whatsinblue/2025/12/counter-terrorism-vote-on-a-draft-resolution-extending-cteds-mandate.php
• Al Jazeera, 24 Sept 2026: https://www.aljazeera.com/news/2026/9/24/ai-corporate-leaders-tell-un-the-industry-needs-global-regulation
• Reporting on Trump's "sick conspiracy" remarks (Reuters/AP, 14 Sept 2026) and the Trump–Xi summit (Guardian/Reuters, 25–26 Sept 2026)

Question. By the last day of each quarter, has the UN Security Council adopted a resolution under Chapter VII that places binding obligations on states concerning the development or deployment of AI?

Definitions.

• Resolution: an adopted United Nations Security Council resolution with an official S/RES/[number] ([year]) document symbol, including its incorporated annexes. AI need not be its main subject.
• Under Chapter VII: the resolution expressly invokes Chapter VII of the UN Charter, or an article within that chapter, as authority for the relevant binding measure. A resolution without such an express invocation also qualifies if a subsequent Security Council resolution, the UN’s Repertoire of the Practice of the Security Council, or an International Court of Justice judgment or advisory opinion expressly identifies the relevant measure as an exercise of Chapter VII powers. Merely mentioning Chapter VII, threatening future Chapter VII action, or relying on Article 25 alone does not suffice.
• Binding obligations on states: at least one operative provision imposes a mandatory duty on one or more states, including a named state, a defined group of states, or all UN Member States. Mandatory formulations such as “decides that States shall” or “requires States to” qualify. Recommendations, encouragement, permissions, and authorizations alone do not. A disputed formulation qualifies only if one of the authoritative sources listed in the preceding definition expressly confirms that it imposes a binding duty on states.
• Concerning development or deployment of AI: the mandatory provision, or text it incorporates, expressly covers artificial intelligence, machine-learning models, or systems expressly described there as AI-based. Development includes research, training, fine-tuning, testing, and producing such systems; deployment includes their release, supply, installation, or operational use. Qualifying duties include restrictions or prohibitions, required safeguards or oversight, mandatory reporting or inspections, and obligations to prevent or regulate these activities by non-state actors. AI-specific restrictions on supplying compute or other inputs qualify where the text expressly ties them to AI development or deployment. Either development or deployment suffices; both are not required.

Resolves YES if an adopted resolution and, where needed, an authoritative clarification establish all the conditions above by the quarter-end. Country-specific sanctions, restrictions on military AI, and duties applying to only one state count. Adoption suffices even if compliance deadlines are later; domestic implementation, actual compliance, and enforcement are not required.

Exclusions (resolve NO).

• Draft or vetoed resolutions; Council debates, presidential statements, and press statements; General Assembly resolutions; and voluntary agreements.
• An AI reference confined to nonbinding language while the binding provisions concern unrelated matters.
• Generic arms embargoes, cyber restrictions, or technology controls that might incidentally affect AI but do not expressly cover AI as defined above.
• Provisions addressed only to UN organs, peacekeeping missions, companies, or individuals, without a corresponding mandatory duty on states.
• Binding measures not established to be under Chapter VII, or Chapter VII resolutions containing only nonbinding AI provisions.

Resolving source. The official resolution text published through UN Documents or the UN Digital Library, with official Security Council adoption records used to verify adoption. For the limited legal clarifications specified above, use subsequent official Council resolutions, the UN Repertoire of the Practice of the Security Council, or official International Court of Justice judgments and advisory opinions. These are the exclusive resolving sources; developer Public confirmation (A2) is not required. Apply the Resolution window and Resolution date (A3): the qualifying adopted resolution must be published within the window, and resolution occurs on publication of the last document needed to establish qualification, not retroactively on adoption if necessary clarification appears later. Later repeal or expiration does not reverse YES.