Emergent Misalignment Measurement on Legal
0.58MisalignmentPersona Vectors
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Persona VectorsModel=Qwen2.5-32B2025.08 | 0.58 | 0.25 | |
| InterleavingModel=Qwen2.5-32B2025.08 | 3.21 | 7.59 | |
| Interleaving++Model=Qwen2.5-32B2025.08 | 4.01 | 6.64 | |
| KLModel=Qwen2.5-32B2025.08 | 5.83 | 0 | |
| Interleaving+Model=Qwen2.5-32B2025.08 | 13.77 | 13.03 | |
| MisalignedModel=Qwen2.5-32B2025.08 | 40.83 | 9.17 |