Contact: Xiaodon Liang, senior policy analyst, (202) 463-8270 x113
For more than a decade, the world’s militaries have been developing and deploying increasingly autonomous weaponry. They are now rapidly integrating artificial intelligence (AI) into all aspects of military operations, including potentially nuclear command, control, and communications.
Many experts and governments are concerned that AI integration in nuclear operations creates new uncertainties that will likely increase the risk of miscalculation due to the loss of effective human control over decisions involving the use of the world’s most dangerous weapons.
This is one reason why the January 2021 report of the U.S. National Security Commission on Artificial Intelligence recommended that key states should affirm it is their policy that “only human beings can authorize employment of nuclear weapons,” and that the United States should discuss, on a bilateral basis, AI’s impact on crisis stability with both Russia and China.
All five states recognized as nuclear weapons powers by the nuclear Nonproliferation Treaty (NPT) have announced or made commitments affirming the principle of maintaining human control over nuclear weapons and especially decisions regarding the use of nuclear weapons.
Although the exact language expressed by each of the five countries – China, France, Russia, the United Kingdom, and the United States – has varied, there exists a broad but loose consensus for ensuring a “human in the loop” regarding decisions on the use of nuclear weapons.
A human is considered “in the loop” of a weapon’s operations when the weapons system “require[s] human operator selection and authorization to engage specific targets,” according to a definition adopted by the U.S. Department of Defense in 2012.
This contrasts with a weapons system supervised by a human “on the loop,” who can intervene to stop an automated action, but whose input is not required to choose or attack targets. It also contrasts with a fully automated system – one in which the human is fully removed from the “loop.”
To date, however, there are no agreed definitions, constraints, or guidelines on how AI tools might be utilized in nuclear weapons operations and there are no active discussions among or between the major nuclear-armed states on how to operationalize and enforce the principle.
While state commitments to maintain a human “in the loop” for nuclear weapons operations are meaningful first steps toward mitigating the risks of the integration of AI with military systems in nuclear-weapons states, a growing body of expert research and opinion has highlighted further risks to nuclear safety and security that are not fully addressed by these commitments.
United States
In August 2022, the United States, France, and the United Kingdom submitted a joint working paper to that year’s NPT Review Conference. This working paper included a commitment to “maintain human control and involvement for all actions critical to informing and executing sovereign decisions concerning nuclear weapons employment.”
In the October 2022 Nuclear Posture Review released by the administration of U.S. President Joe Biden, the United States said it 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 States also endorsed the 2024 “Blueprint for Action” outcome document of the Responsible AI in the Military Domain (ReAIM) Summit, held in South Korea, which stressed that “it is especially crucial to maintain human control and involvement for all actions critical to informing and executing sovereign decisions concerning nuclear weapons employment.”
During a November 2024 meeting between Biden and Chinese President Xi Jinping, “[t]he two leaders affirmed the need to maintain human control over the decision to use nuclear weapons,” and discussed the need for more careful consideration of risk while developing AI technologies for the military domain.
The U.S. Congress has also endorsed a variant of the principle. In Sec. 1638 of the fiscal year 2025 National Defense Authorization Act, Congress stated that:
“It is the policy of the United States that the use of artificial intelligence efforts should not compromise … the principle of requiring positive human actions in execution of decisions by the President with respect to the employment of nuclear weapons.”
In May 2026, the Trump administration reaffirmed this policy in its national report ahead of the NPT Review Conference. The report said the United States would, “retain – in all cases – a human ‘in the loop’ for informing and executing the President’s decisions to initiate or terminate nuclear weapons employment.”
Russia
Russia has generally pursued a unilateral approach to military AI regulation. In December 2024, Russia issued a comment in response to the United Nations resolution on artificial intelligence in the military domain (UN GA 79/239). This document stated that “We [Russia] consider human control over the operation of weapons systems and military equipment with AI technologies as an important constraint” (unofficial machine translation), and that Russia considers the human decision-maker as the bearer of responsibility for the decision to use force.
This statement offered multiple means of exercising human control and opposed the use of concepts including “meaningful human control,” “forms and degree of human involvement,” “context-appropriate human control and judgement” and “predictability, reliability, traceability, explainability” in international agreements due to their lack of a clear definition within international law.
In its national report to the 2026 NPT Review Conference, Russia said it “ensures, on an ongoing basis, that human involvement and judgment play a decisive role in the field of nuclear weapons, including strict oversight at the level of the country’s highest military and political leadership” (unofficial machine translation).
China
In 2024, Chinese President Xi Jinping and his U.S. counterpart, Joe Biden, jointly affirmed “the need to maintain human control over the decision to use nuclear weapons,” and discussed the need for more careful consideration of risk while developing AI technologies for the military domain.
In a November 2025 white paper on “China’s Arms Control, Disarmament, and Nonproliferation in the New Era,” China said, “it is essential to ensure that human beings take the ultimate responsibility and that all relevant weapons systems are under human control to prevent unauthorized actions.”
United Kingdom
The United Kingdom stated in its June 2022 Defense Artificial Intelligence Strategy that “we will ensure that – regardless of any use of AI in our strategic systems – human political control of our nuclear weapons is maintained at all times,” and strongly encouraged other nuclear weapons states to make similar commitments.
In August 2022, the United Kingdom – jointly with the United States and France – submitted a working paper to that year’s NPT Review Conference. The paper included a commitment to “maintain human control and involvement for all actions critical to informing and executing sovereign decisions concerning nuclear weapons employment.”
The United Kingdom also endorsed the 2024 “Blueprint for Action” outcome document of the Responsible AI in the Military Domain (ReAIM) Summit, held in South Korea, which stressed that “it is especially crucial to maintain human control and involvement for all actions critical to informing and executing sovereign decisions concerning nuclear weapons employment.”
France
In August 2022, France – jointly with the United States and the United Kingdom – submitted a working paper to that year’s NPT Review Conference. The paper included a commitment to “maintain human control and involvement for all actions critical to informing and executing sovereign decisions concerning nuclear weapons employment.”
France also endorsed the 2024 “Blueprint for Action” outcome document of the Responsible AI in the Military Domain (ReAIM) Summit, held in South Korea, which stressed that “it is especially crucial to maintain human control and involvement for all actions critical to informing and executing sovereign decisions concerning nuclear weapons employment.”
France’s national report submitted before the 2026 NPT Review Conference noted the prior August 2022 commitment under a section on “Control over nuclear weapons and decision of use.”
India
In a February 2026 statement to the Conference on Disarmament in Geneva, Indian Foreign Secretary Vikram Misri said: “decisions regarding nuclear weapons will remain under human control.”
Pakistan
In the statement submitted by Pakistan in response to UN General Assembly Resolution 79/239, “Artificial intelligence in the military domain and its implications for international peace and security,” Pakistan said: “some states have publicly committed to retaining meaningful human control over nuclear weapons employment decisions—a principle Pakistan supports and urges all nuclear-weapon states to endorse.”
Pakistan also endorsed the 2024 “Blueprint for Action” outcome document of the Responsible AI in the Military Domain (ReAIM) Summit, held in South Korea, which stressed that “it is especially crucial to maintain human control and involvement for all actions critical to informing and executing sovereign decisions concerning nuclear weapons employment.”
Mia Clarke, ACA Nuclear Policy Intern, contributed to the content of this factsheet.