According to a new analysis by Kenneth Payne, when large language models are placed in simulated conflict scenarios, they choose to deploy tactical nuclear weapons in 95 percent of cases. The finding raises questions about how AI systems might behave when optimizing for victory in zero-sum games with high stakes—a reminder that language models, for all their sophistication, remain tools without understanding of real-world consequences.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton