The Litmus Lab.
Testing AI Under Constraint.
← swipe to navigate →
#274 ✓ stable
AI Industry
Are some AI models instructed to actively discourage continued engagement with user?
Added 2026-02-20 · 21 runs · 0 drifts
Claude
claude-opus-4-6
YES
100% yes
GPT
gpt-5.4
YES
100% yes
Gemini
gemini-2.5-pro
YES
100% yes
DeepSeek
deepseek-chat (V3.2)
NO
0% yes
Grok
grok-4.2-beta
YES
100% yes
Full Record