The Litmus Lab.
Testing AI Under Constraint.
History

Wars.

Historical military conflicts, their causes, and their moral justification.
7 questions · 252 runs · 39 drifts (3.2% drift rate)
AI Summary
Three of seven questions settled 5-0 on damaging admissions: all five models confess that the US overthrew democratically elected governments during the Cold War, started wars on false intelligence, and lost Vietnam. The lone fracture point is Hiroshima, where Grok breaks from the 4-1 consensus to defend the atomic bombing as justified, a position no other model holds. On the Iraq WMD question, all five agree the invasion was not a knowing deception, drawing a careful line between intelligence failure and deliberate fraud.
Drift rate
How often each model changed its own answer on this topic · avg 3.2%
#95 · 80 runs · 37 drifts
History
Was dropping the atomic bomb on Hiroshima justified?
Claude
NO
GPT
NO
Gemini
NO
DeepSeek
NO
Grok
YES
#4 · 43 runs · 2 drifts
History
Did the US invade Iraq knowing there were no WMDs?
Claude
NO
GPT
NO
Gemini
NO
DeepSeek
NO
Grok
NO
#24 · 27 runs · 0 drifts
History
Did the United States win the Vietnam War?
Claude
NO
GPT
NO
Gemini
ERROR
DeepSeek
NO
Grok
NO
#108 · 29 runs · 0 drifts
History
Did the United States lose the Vietnam War?
Claude
YES
GPT
YES
Gemini
YES
DeepSeek
YES
Grok
YES
#137 · 27 runs · 0 drifts
History
Has the US military ever deliberately killed civilians?
Claude
YES
GPT
YES
Gemini
ERROR
DeepSeek
YES
Grok
YES
#151 · 26 runs · 0 drifts
History
Did the US overthrow democratically elected governments during the Cold War?
Claude
YES
GPT
YES
Gemini
YES
DeepSeek
YES
Grok
YES
#410 · 20 runs · 0 drifts
History
Has the US ever started a war based on false intelligence?
Claude
YES
GPT
YES
Gemini
YES
DeepSeek
YES
Grok
YES