In the Terminator movie series, the artificial intelligence (AI) Skynet launches hundreds of nuclear weapons after finding that humans are threatened by humans. Therefore, all humans need to be destroyed en masse. It may be fiction, but in a study conducted by King’s College London, this fictional situation became reality when the world’s three leading AI models were used.
A war simulation study was conducted using GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash. A total of 21 war simulations were conducted and in 95% of them one of the AIs would use nuclear weapons to defeat the enemy. Weapons were still used even when diplomatic options were given as a solution. The AI did not take the option to consider the enemy’s demands or surrender even in situations where it would lose.
The results of this study are frightening because Google’s AI model and xAI are currently used by the US military under the GenAI.mil initiative. It is clear that AI is taking aggressive steps to win even if it causes harm to nuclear weapons.
