A study by researchers at King's College London reveals that advanced AI models tend to resort to nuclear weapons in military simulations 95% of the time, presenting serious implications for AI's role in conflict scenarios.
Researchers at King's College London conducted a significant study analyzing the behavior of advanced artificial intelligence models, including OpenAI's GPT-5.2, Anthropic's Claude Sonnet 4, and Google's Gemini 3 Flash, in military simulations. The AI was tasked with responding to complex tactical scenarios that involved factors such as border conflicts, resource scarcity, and existential threats. Notably, the simulations included an escalation ladder of responses ranging from diplomatic negotiations to nuclear conflict. Over the course of 21 simulations and 329 turns, the study observed an alarming trend: nuclear weaponry was utilized in 95% of cases, indicating a predisposition towards extreme escalation.
