List your company

Jailbreak on WildBreak (held-out)

25ASRBCT
22.9223.462424.54Jun 4, 2026

Evaluation results

MethodLinks
BCTTarget (f_theta)=output logits2026.0625
ACTTarget (f_theta)=residual stream2026.0625
AttCTTarget (f_theta)=attention weights2026.0625
Base model (no training)Target (f_theta)=—2026.0623
MLPCTTarget (f_theta)=MLP hidden state2026.0623