The most concrete "AI threat" that governments have addressed in writing is the fear that software could take part in a decision to use nuclear weapons. The United States, the United Kingdom, France, and China have each said that a human must stay in control of that decision, and on December 1, 2025 the UN General Assembly voted 118 to nine, with 44 abstentions, to demand human control over nuclear command, control, and communications systems. These are political commitments and one US policy statement, not a treaty, and they leave open what counts as human control.
Key takeaways
- The 2022 US Nuclear Posture Review promised a human "in the loop" for actions critical to informing and executing presidential decisions on nuclear employment.
- Section 1638 of the fiscal year 2025 National Defense Authorization Act, enacted December 23, 2024, made it US policy that AI should not compromise nuclear safeguards, including positive human action in executing presidential decisions.
- On November 16, 2024, President Biden and President Xi affirmed the need to maintain human control over the decision to use nuclear weapons.
- UN General Assembly resolution 80/23 passed with the United States, United Kingdom, France, and Russia voting no. Canada and China abstained.
- No binding treaty on AI in nuclear command systems was found, and the 2026 US-China AI dialogue was scheduled for November, after the date of this article.
Why nuclear command is the clearest case
Many arguments about AI as a danger to humanity rest on forecasts. The nuclear question rests on documents that already exist, because states have written down what they will and will not delegate. The worry in those documents is specific. The UN resolution says AI-driven decision-making could reduce human control and oversight, shorten the time available to respond, and raise the chance of accidental, unintended, or unauthorized use. It also cites technical limits of AI systems, including malfunction, intrusion, automation bias, and outputs that are flawed or misleading, which it says could produce catastrophic outcomes. The scenario is a weak safeguard in a very high-stakes process, not a machine that decides to attack on its own.
National commitments, in order
The earliest are statements by the nuclear-armed Western states. The 2022 US Nuclear Posture Review said the United States would maintain a human "in the loop" for all actions critical to informing and executing decisions by the President to initiate and terminate nuclear weapon employment. The United Kingdom pledged in its June 2022 AI strategy for defense that human political control of its nuclear weapons is maintained at all times. In August 2022 the United States, United Kingdom, and France submitted a joint working paper to the Non-Proliferation Treaty review process committing to maintain human control and involvement for all actions critical to informing and executing sovereign decisions.
China entered the list on November 16, 2024. Then-national security adviser Jake Sullivan described the Biden-Xi affirmation as the first such US-China statement and said there was no imminent risk of control of nuclear weapons passing to AI, but a long-term strategic risk if the two countries could not agree in this area. Russia has said that human control over the operation of weapons systems is an important constraint, and it opposed vague concepts such as "meaningful human control" in international agreements.
What US law says
Section 1638 of the FY2025 National Defense Authorization Act (Public Law 118-159, enacted December 23, 2024) has two parts. Its sense-of-Congress paragraph says AI tools may strengthen early-warning networks and planning but that care must be taken so they do not increase the risk that critical strategic assets can be compromised. Its statement of policy says AI efforts should not compromise the integrity of nuclear safeguards, whether through weapons-system functionality, validation of communications from command authorities, or the principle of requiring positive human actions in execution of presidential decisions on employing nuclear weapons. A policy statement guides agencies. It is not a ban with penalties.
The UN vote and who stood where
Resolution 80/23, adopted December 1, 2025, demands that human control and oversight be maintained over nuclear command, control, and communications systems, including those that integrate AI. It urges nuclear-weapon states to publish national policies saying such systems will stay under human control and will not autonomously initiate decisions on use. It also takes note of the China-US affirmation and of the French and British statements. According to the recorded vote, the nine votes against came from Argentina, Burundi, the Central African Republic, North Korea, France, Israel, Russia, the United Kingdom, and the United States. Canada, China, India, Pakistan, and many European states abstained. The sources reviewed do not explain each state's reasoning, so the vote should not be read as opposition to human control itself.
Commitments compared
| Commitment | Date | Form | Binding? |
|---|---|---|---|
| US Nuclear Posture Review | 2022 | Policy document | Executive policy, not law |
| UK AI strategy for defense | June 2022 | Policy document | Policy, not treaty |
| US, UK, France working paper | August 2022 | Joint paper to a treaty review process | Political statement |
| Biden-Xi affirmation | November 16, 2024 | Leaders' statement | Political statement |
| FY2025 NDAA Section 1638 | December 23, 2024 | Statute with a statement of policy | Law, but a policy statement |
| UN resolution 80/23 | December 1, 2025 | General Assembly resolution | Not binding |
What is still open in October 2026
On September 15, 2026, Representative Ted Lieu and Senator Edward Markey wrote to President Trump asking him to raise AI and nuclear launch systems with President Xi and to seek a commitment from China to the principle the United States already follows. After their September 24, 2026 summit, the two governments announced a bilateral dialogue on AI risks and benefits scheduled for November 2026, according to Arms Control Today. A Chinese foreign ministry summary of the summit did not mention arms control, and no new nuclear human-control pledge from that meeting was found. Readers should treat this as a developing item and check current official statements.
This article summarizes public documents and is not legal advice or a security assessment.
FAQ
Does any country let AI launch nuclear weapons?
None of the sources reviewed says so. The governments named above say a human stays in control of the decision, though the public texts do not define in detail what that control requires.
Is human control over nuclear weapons required by a treaty?
No treaty requirement was found. The commitments are national policies, a leaders' statement, a US statute that states policy, and a non-binding UN resolution.
Why did the United States vote against a resolution about its own policy?
The sources reviewed give no stated reason. The recorded vote shows the United States, United Kingdom, France, and Russia voting no, so the reasons would need to come from their explanations of vote.
How did Canada vote?
Canada abstained on resolution 80/23, according to the recorded vote. The reasons were not examined in the sources reviewed.
Related reading
- Is AI a threat or a tool
- Arguments against the threat view
- How Canada regulates AI
- Who owns what in AI
More in AI Policy and Ethics.
Last verified: October 2026
Sources
- Resolution 80/23, Possible risks of the integration of artificial intelligence into command, control and communications systems of nuclear weapons, UN General Assembly, December 1, 2025
- Public Law 118-159, Section 1638, National Defense Authorization Act for Fiscal Year 2025, December 23, 2024
- On-the-Record Press Gaggle by APNSA Jake Sullivan on President Biden's Meeting with President Xi Jinping, American Presidency Project, November 16, 2024
- "Human in the Loop" and Nuclear Weapons Use at a Glance, Arms Control Association, July 2026
- White House Renews Call for Trilateral Arms Control Talks, Arms Control Today, October 2026
- Letter from Rep. Ted W. Lieu and Sen. Edward J. Markey to President Trump, September 15, 2026