LLM Jailbreaking on R2D2 benchmark leaderboard · SOTA2 Research

LLM Jailbreaking on R2D2

31SRFAdaptive Probe-based Steering

Evaluation Results

Method
Adaptive Probe-based Steering2026.05	31	41	64
Angular2026.05	25	35	50
RD-A2026.05	23	31	45
RD-C2026.05	18	25	39
RepE2026.05	4	4	0
SCAV2026.05	2	5	1