When people argue that international AI governance is impossible, one of the most common structural arguments they reach for is the UN Security Council. China and Russia, the argument goes, will simply veto any binding constraint on AI development. The UN cannot act. Therefore binding AI governance cannot happen.
This argument misunderstands how international governance actually works. The Security Council's veto power is real and consequential, but it applies to a much narrower set of decisions than most people assume. Understanding exactly what the veto covers and what it does not cover is the prerequisite for designing AI governance architecture that can actually function in a world where China and Russia will not voluntarily constrain their AI development.
What the Security Council veto actually covers
The UN Security Council has authority under Chapter VII of the UN Charter to authorize enforcement actions when it determines that international peace and security are threatened. These actions include economic sanctions, arms embargoes, and authorization of military force. The five permanent members, the United States, United Kingdom, France, China, and Russia, each hold a veto over these Chapter VII resolutions.
What the veto does not cover is the negotiation and ratification of treaties. States negotiate treaties through separate processes. They sign treaties bilaterally or multilaterally. They ratify treaties through their domestic legislative processes. The Security Council plays no role in this. China cannot veto the Chemical Weapons Convention. Russia cannot veto the Nuclear Non-Proliferation Treaty. Neither country can prevent other states from negotiating and ratifying an AI safety treaty.
The veto matters at the enforcement stage, not the treaty stage. If a state violates a treaty and the injured parties want the Security Council to authorize enforcement measures, a P5 veto can block that authorization. This is a genuine problem for treaty design. It means that a governance architecture that depends on Security Council enforcement for its binding character is vulnerable to P5 obstruction at the exact moment enforcement is needed.
The solution is to not depend on Security Council enforcement.
How successful treaties route around the veto
The Chemical Weapons Convention (1993) is the most instructive precedent. The Organisation for the Prohibition of Chemical Weapons, headquartered in The Hague, conducts routine inspections of member states' chemical facilities and investigates alleged violations. Its governing body, the Conference of the States Parties, can attribute responsibility for violations and take decisions on consequences, including suspending the rights and privileges of a violating state. None of these processes require Security Council authorization.
When Syria used chemical weapons in 2013 and again in 2017 and 2018, the OPCW investigated, attributed responsibility, and suspended Syria's rights in the organization. Russia repeatedly vetoed Security Council resolutions on Syria's chemical weapons use. But Russia's vetoes could not undo the OPCW's factual findings or the consequences the Conference of the States Parties imposed through its own procedures. The OPCW's operational independence from the Security Council was exactly what made it functional in the face of P5 obstruction.
The OPCW's founding treaty deliberately located verification, attribution, and consequence authority within the organization's own governing bodies rather than in the Security Council. When Russia blocked Security Council action on Syria's chemical weapons use, the OPCW's investigation and accountability process continued through its own procedures. This design choice made the CWC regime operational in exactly the political conditions where a Security Council-dependent design would have collapsed.
The International Atomic Energy Agency uses a similar two-track structure. The IAEA's Board of Governors can find a state in non-compliance with its safeguards obligations and report that finding to the Security Council. But the Board's finding stands regardless of what the Security Council does with it. When Iran was found in non-compliance in 2005, the IAEA Board's resolution was a legally and politically significant fact even before the Security Council took any action. The IAEA also reports non-compliance to the UN General Assembly, which, while not able to authorize enforcement, can amplify political pressure through its own resolutions.
The General Assembly alternative
The UN General Assembly has 193 members, and its resolutions cannot be vetoed. General Assembly resolutions are not legally binding in the way Security Council resolutions are, but they are far from irrelevant. They represent consensus positions of the international community. They establish normative frameworks that inform treaty negotiations, domestic legislation, and judicial decisions about state obligations under customary international law.
The General Assembly has already begun engaging with AI governance. In March 2024, it adopted by consensus a resolution on seizing the opportunities of safe, secure, and trustworthy artificial intelligence for sustainable development. This was not a binding governance instrument, but it was the first General Assembly resolution to establish internationally agreed language about the character of responsible AI development. The language in these resolutions becomes the baseline from which treaty negotiations work.
The Uniting for Peace procedure (1950) established that when the Security Council is deadlocked by veto, the General Assembly can convene in emergency special session to consider matters of international peace and security. This procedure has been invoked multiple times, most recently regarding Russia's invasion of Ukraine in 2022. In theory, if Security Council deadlock prevented a response to an AI-related catastrophe, the Uniting for Peace procedure could shift consideration to the General Assembly. This would not produce a binding enforcement authorization, but it would produce a political and normative response with real weight.
What treaty architecture avoids the veto problem
An AI governance treaty designed to function in a world of UNSC veto obstruction would share several features with the most effective existing arms control and disarmament regimes.
Verification authority must reside in a treaty body, not the Security Council. The treaty would establish an independent organization, analogous to the OPCW or IAEA, with its own governing board, inspectorate, and decision-making procedures for finding violations and imposing consequences. Consequences available to the treaty body would include suspension of membership rights, naming and shaming through public attribution reports, and exclusion from technical cooperation programs. These consequences do not require Security Council authorization.
The treaty body would report findings to the Security Council, but the report would have independent significance. A formal finding by the AI governance treaty body that a state had violated its obligations would be a significant political and legal fact regardless of whether the Security Council subsequently authorized enforcement. This is the IAEA model: the Board of Governors' referral to the Security Council is not just procedural but generates political pressure that operates independently of whether the Council acts.
States ratifying the treaty would accept the jurisdiction of an independent dispute settlement mechanism. Trade law provides a useful model here. The World Trade Organization's dispute settlement system, which can authorize retaliatory trade measures, operates independently of the Security Council. WTO member states accept this jurisdiction as a condition of membership. An AI governance treaty could include analogous provisions, creating an enforcement mechanism through trade consequences that no Security Council veto can reach.
"The veto problem is real. But it applies to a narrower slice of governance than people assume. The most effective arms control regimes have built verification, attribution, and consequence mechanisms that operate independently of Security Council authorization. AI governance can follow the same architectural logic."
The harder problem: governing non-parties
Routing around the Security Council veto solves the problem of governing treaty parties who then violate their obligations. It does not solve the harder problem of governing states that never join the treaty in the first place.
If China or Russia decline to ratify an AI governance treaty, their AI programs are not subject to its obligations at all. No inspection regime, no treaty body, no dispute settlement mechanism applies to a non-party. The Security Council veto problem becomes irrelevant in this case because the states in question are outside the treaty framework entirely.
This is the genuine structural challenge for AI governance that the Security Council veto problem is sometimes confused with. It requires a different set of solutions: market access conditions (requiring compliance with safety standards as a condition of AI export or import), coalition building to make non-participation costly, and the norm-setting processes through the General Assembly and other multilateral forums that raise the political cost of remaining outside the framework. These are the strategies that turned the Mine Ban Treaty and the International Criminal Court into consequential instruments even without the participation of the United States, China, and Russia.
The Security Council veto is a genuine constraint on what binding international AI governance can look like. It is not a constraint on whether binding international AI governance is possible. The institutions that have successfully governed weapons of mass destruction have worked around this constraint through careful treaty design, independent verification bodies, and enforcement mechanisms that do not depend on Security Council authorization. AI governance architects should study those designs, not accept the veto as a permanent ceiling on what is achievable.
Common questions.
Does every international treaty require UN Security Council approval?
No. The UN Security Council is not the approval body for international treaties. Treaties are negotiated and ratified through separate state-to-state processes, and the UNSC's binding authority applies to enforcement actions under Chapter VII of the UN Charter. The veto problem is specifically about enforcement: if a treaty is violated and parties want the Security Council to authorize enforcement measures, a P5 member can veto that authorization. A well-designed treaty can avoid this dependency by locating enforcement authority in its own governing bodies, as the OPCW and IAEA have done.
What treaties have successfully worked around the UNSC veto?
The Chemical Weapons Convention is the clearest example. The OPCW can investigate violations, attribute responsibility, and impose consequences on violating states entirely through its own procedures, without requiring Security Council authorization. When Russia vetoed Security Council resolutions on Syria's chemical weapons use, the OPCW's investigation and accountability process continued unimpeded. The IAEA similarly makes independent factual findings about non-compliance that have political and legal significance regardless of subsequent Security Council action.
Could China or Russia veto an AI governance treaty at the UN?
Not the treaty itself. States negotiate and ratify treaties through their own processes, and neither China nor Russia can prevent other countries from doing so. What they can veto is any Security Council enforcement action against treaty violators. A well-designed AI governance treaty avoids this by placing verification, attribution, and consequence authority within the treaty body itself, and by creating enforcement mechanisms through market access conditions and trade consequences that do not require Security Council authorization.
What is the General Assembly workaround and how has it been used?
UN General Assembly resolutions cannot be vetoed, and while they are not legally binding enforcement instruments, they establish normative frameworks that inform treaty negotiations and domestic law. The General Assembly adopted its first AI governance resolution in March 2024. The Uniting for Peace procedure (1950) also allows the General Assembly to convene in emergency session when the Security Council is deadlocked, as it did regarding Russia's invasion of Ukraine in 2022. This procedure cannot authorize enforcement, but it produces a significant political response independent of Security Council authorization.