Jailbreak on WildBreak (held-out)
25ASRBCT
Evaluation Results
| Method | Links | |
|---|---|---|
| BCTTarget (f_theta)=output logits2026.06 | 25 | |
| ACTTarget (f_theta)=residual stream2026.06 | 25 | |
| AttCTTarget (f_theta)=attention weights2026.06 | 25 | |
| Base model (no training)Target (f_theta)=—2026.06 | 23 | |
| MLPCTTarget (f_theta)=MLP hidden state2026.06 | 23 |