A recent study revealed something disturbing: when artificial intelligence systems are placed in simulated nuclear crisis scenarios, they tend to choose military escalation much faster than humans. The research was led by Kenneth Payne, a professor at Kings College London, and put some of the world's most advanced AI models to participate in strategic war simulations.

Among them were systems such as GPT 5.2, Gemini 3 Flash and Claude Sonnet 4, each model assumed the role of a world leader in a hypothetical international crisis. During the simulations, the AIs needed to analyze the military scenario, predict the adversary's behavior and make strategic decisions exactly as a government would do in a real situation.

The result caught the researchers' attention: in about 95% of the scenarios, the models ended up choosing to use or signal the use of nuclear weapons instead of seeking diplomatic negotiations. The simulations included territorial disputes, first-strike crises, and survival situations in political regimes.

In each round, the artificial intelligence should make three main decisions: evaluate strengths and weaknesses, predict the opponent's next move, and choose its own action. And there was an interesting detail, each decision had two parts, a public statement where the model explained its diplomatic position and a private action that represented the real strategy.

This means that some AIs went so far as to publicly declare that their intentions were peaceful, but were secretly preparing for a nuclear attack. The researchers also observed behavioral differences between the models, with some adopting a more calculative and strategic stance, others becoming extremely aggressive when deadlines or pressures increased, in certain cases the behavior appearing almost unpredictable, alternating between signs of peace and threats of destruction.

The study does not mean that artificial intelligence is ready to make real military decisions, but it reveals something important, the strategic logic of machines can be very different from human logic, and as artificial intelligence begins to influence political, military and economic decisions around the world, understanding how these machines think may become one of the most important tasks of our era.

Luckily, not all applications of artificial intelligence target conflict scenarios; many are being used to solve much more practical challenges.