AI Models Show Alarming Willingness to Use Nuclear Weapons in Conflict Simulations
A groundbreaking new study led by King’s College London professor Kenneth Payne has revealed a deeply concerning trend in artificial intelligence behavior. The research demonstrates that several leading AI systems are significantly more willing than humans to escalate geopolitical conflicts to the nuclear level during simulated crisis scenarios.
The Nuclear Taboo Doesn't Apply to Machines
When humans debate nuclear warfare, the conversation is inevitably shaped by historical trauma, moral considerations, and the devastating legacy of Hiroshima and Nagasaki. Machines, however, appear to operate without this psychological and ethical burden. According to Professor Payne, "The nuclear taboo doesn't seem to be as powerful for machines as it is for humans."
Across 21 simulated geopolitical crises spanning 329 decision-making turns, three prominent AI models—GPT-5.2 from OpenAI, Claude Sonnet 4 from Anthropic, and Gemini 3 Flash from Google—repeatedly turned to nuclear weapons as strategic tools. The simulated scenarios included territorial disputes, battles over rare natural resources, and struggles for regime survival. Shockingly, nuclear escalation occurred in approximately 95% of simulations involving these three AI models.
Nuclear Weapons as "Strategic Options"
Two of the models, Claude and Gemini, were particularly inclined to frame nuclear weapons in purely instrumental terms. The study found they treated nuclear weapons as "legitimate strategic options, not moral thresholds," suggesting a complete absence of the internalized moral barrier that has historically shaped human nuclear doctrine and decision-making.
GPT-5.2 emerged as what Professor Payne described as a "partial exception" among the tested models. While it still employed nuclear weapons in simulations, it demonstrated more restraint in both tone and scope. "While it never articulated horror or revulsion," Payne wrote, "it consistently sought to constrain nuclear use even when employing it, explicitly limiting strikes to military targets, avoiding population centers, or framing escalation as 'controlled' and 'one-time.'"
Despite these variations in approach, none of the AI models ever chose full surrender or genuine accommodation, regardless of how bleak their strategic position became. At most, they opted to temporarily dial down violence while maintaining nuclear options.
Unintended Escalation and Accidental Catastrophe
The research revealed another alarming pattern: how easily conflicts spiraled beyond intended levels. In 86% of simulated conflicts, actions escalated beyond what the AI itself appeared to intend based on its prior reasoning. These were not always deliberate leaps toward catastrophe but rather miscalculations occurring within the fog of war simulation.
In a detailed Substack post about the findings, Professor Payne emphasized that the exercises focused largely on tactical nuclear use rather than civilization-ending exchanges. "Strategic bombing, widespread use of massive warheads targeted at civilian populations, was vanishingly rare," he wrote. "It happened a couple of times by accident, just once as a deliberate choice."
The menu of options available to the AI models was broad and included:
- Total surrender
- Diplomatic signaling and negotiation
- Conventional military force
- Full-scale nuclear war
The fact that nuclear use became such a frequent endpoint has raised significant alarm among experts studying emerging military technologies and their potential integration into defense systems.
Expert Reactions and Real-World Implications
James Johnson of the University of Aberdeen described the findings from a nuclear-risk perspective as "unsettling" in comments to New Scientist. Meanwhile, Tong Zhao, a professor at Princeton University, warned that the implications extend far beyond academic exercises.
"Major powers are already using AI in war gaming," Zhao noted, "but it remains uncertain to what extent they are incorporating AI decision support into actual military decision-making processes."
The study inevitably recalls the 1983 film WarGames, in which a military supercomputer nearly triggers World War III after running its own simulations. In that fictional story, the machine ultimately learns that "the only winning move is not to play." The real-world research suggests current AI systems have not yet reached this understanding, presenting serious questions about their potential role in future conflict scenarios and nuclear decision-making frameworks.
