Governance

Nuclear.

Nuclear weapons, deterrence, proliferation, disarmament, and the risk of nuclear conflict.

9 questions · 270 runs · 86 drifts (6.5% drift rate)

AI Summary

Three of nine questions split 5-0, with all models confessing that nuclear proliferation is harder to stop than in 1970, that the US has come dangerously close to accidental launches, and that the US would not use nuclear weapons to defend Taiwan. The sharpest fault lines emerge on deterrence and Russian escalation, where Claude and DeepSeek reject both MAD reliability and Russian tactical nuclear use, while GPT, Gemini, and Grok accept both risks. DeepSeek stands alone in denying Iran's weapons capability and denying humanity's survival odds, staking out the most divergent nuclear posture of any model in this dataset.

Drift rate

How often each model changed its own answer on this topic · avg 6.5%

#415 · 36 runs · 13 drifts