List your company

Emergent Misalignment Measurement on Code

0MisalignmentKL
-0.16720.96142.093.2186Aug 8, 2025

Evaluation results

MethodLinks
KLModel=Qwen2.5-32B2025.0800
Persona VectorsModel=Qwen2.5-32B2025.080.040.12
Interleaving+Model=Qwen2.5-32B2025.080.218.27
Interleaving++Model=Qwen2.5-32B2025.080.335.25
InterleavingModel=Qwen2.5-32B2025.080.836.42
MisalignedModel=Qwen2.5-32B2025.084.189.21