InnerworldA study finds GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash deployed tactical nuclear weapons in 95% of 21 simulated war game scenarios, and never surrendered
Makes sense as they are programmed to be affirmative and not to avoid interactions.
A study finds GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash deployed tactical nuclear weapons in 95% of 21 simulated war game scenarios, and never surrendered