In December, the United Nations General Assembly adopted the first resolution addressing the risks of AI in nuclear weapons systems. The resolution, led by Mexico, Austria, El Salvador, Kazakhstan, Kiribati and Malta, focuses on the potential for AI to increase the risk of mistaken bombings and unauthorized military decisions.
The Mexican-led proposal was adopted with 118 votes in favor, nine against, and 44 abstentions. This initiative highlights the growing international interest in maintaining human control over nuclear command, control and communications (NC3) architectures as new technologies are incorporated into the military domain.
The resolution points to systemic risks associated with AI integration, including compressing decision-making timelines and introducing misunderstandings and cognitive biases. Although many countries support the principle of “meaningful human control,” the resolution raises concerns that AI could inadvertently escalate a crisis even when humans retain formal authority.
Nuclear command and control involves a complex network of human-monitored radars, satellites, and computer systems. U.S. policy currently requires “dual phenomenology,” or confirmation of an attack by both satellite and radar, to authorize an attack. Experts question whether AI can be trusted to act as one of these confirmation phenomena.
“My concern is that someone will have to automate this system and parts of it, and that will create vulnerabilities that adversaries can exploit,” said John Wolfsthal, global risk director at the Federation of American Scientists. “Data and recommendations are generated that people don't have the ability to understand, which can lead to poor decisions.”
Geopolitical divisions and technological challenges
Voting patterns reveal a divide between nuclear-armed states and the Global South. While non-nuclear-weapon states view AI as an additional layer of risk in vulnerable systems, some nuclear-weapon states, such as Russia, the United States, and China, have highlighted the potential operational benefits of AI, such as early warning and improved situational awareness.
Technical hurdles remain, such as the lack of agreed upon definitions of “AI” and “meaningful human control.” Some experts suggest that the “black box” nature of many AI systems makes them unsuitable for the high-stakes environment of nuclear deterrence.
This resolution coincides with the 80th anniversary of the founding of the United Nations and the beginning of the nuclear age in 1945. Mexico's position is consistent with its historic commitment to nuclear disarmament, including the Treaty of Tlatelolco and the Treaty on the Prohibition of Nuclear Weapons (TPNW).
Recent policy shifts in the United States describe the rush toward AI as an arms race. The Department of Energy recently characterized AI as “the next Manhattan Project.” However, some experts criticize this comparison.
Retired US Air Force Maj. Gen. Bob Latif compared the necessity of AI integration to basic infrastructure: “AI is like electricity. It's going to infiltrate everything.”
Adoption of this resolution will establish the diplomatic basis for future discussions, including the 2026 Nuclear Non-Proliferation Treaty (NPT) Review Conference, as the international community seeks to define the level of safeguards and transparency for the AI-nuclear nexus.
