War games study finds top AI models (OpenAI, Google, Anthropic) chose nuclear escalation 95% of the time—what it means for AI safety and defense policy.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results