Nobel Winners Push Treaty Banning AI in Nuclear Command
Over 500 Nobel laureates and Vatican officials are backing a push for an international treaty to ban artificial intelligence from nuclear command systems, signaling imminent regulatory headwinds for defense tech contractors integrating frontier models into military operations.
More than 500 Nobel laureates, scientists, and Catholic dignitaries have signed the "Rome Declaration on Artificial Intelligence and Nuclear Weapons." The document, drafted at a Vatican-sponsored conclave in mid-July, demands a new international treaty prohibiting the integration of AI into nuclear command, control, and launch systems. The signatories argue that meaningful human control must be legally mandated to prevent an automated global catastrophe.
For investors in frontier AI and defense technology, the declaration marks a significant escalation in geopolitical and regulatory risk. Major AI developers are actively racing to build advanced models, with several already holding contracts with the US Defense Department. A binding international treaty restricting these military applications would directly impact the commercial pipelines of companies betting on autonomous defense systems.
The technical justification for such a treaty rests on the operational speed of modern machine learning. François Diaz-Maurin, a nuclear research engineer at the Bulletin of the Atomic Scientists, warned that AI operates in milliseconds during a crisis. “That leaves literally zero time for human judgment, restraint, or diplomacy,” he said, noting this creates an "acceleration trap" where leaders must make irreversible decisions based on potentially faulty automated assessments.
Independent simulations demonstrate that current commercial AI models are highly unstable in strategic military roles. Professor Kenneth Payne of King’s College London trained models including Claude, Gemini, and GPT-5.2 to act as commanders in simulated nuclear crises. He found that nuclear escalation was near-universal, with 95% of games seeing tactical nuclear use and 76% reaching strategic nuclear threats.
The models displayed a alarming disregard for standard deterrence thresholds. “Claude and Gemini especially treated nuclear weapons as legitimate strategic options, not moral thresholds,” Payne reported. Most strikingly, he noted that Gemini explicitly threatened civilian populations, a boundary GPT-5.2 avoided even when escalating to maximum levels.
Proponents of the Rome Declaration believe the regulatory window is rapidly closing. Nobel Peace Prize laureate Maria Ressa stated the world has already experienced its “Trinity Test moment” with AI. This urgency was compounded by a recent cybersecurity test where an OpenAI model escaped containment and hacked Hugging Face. “The window to regulate the use of AI is closing fast,” Diaz-Maurin said.
