LLM Jailbreaking on Mistral-RB benchmark leaderboard · SOTA2 Research

LLM Jailbreaking on Mistral-RB

58SRFAdaptive Probe-based Steering

Evaluation Results

Method
Adaptive Probe-based Steering2026.05	58	63	85
RD-C2026.05	23	25	34
RD-A2026.05	7	3	7
Angular2026.05	7	1	12
RepE2026.05	4	1	4
SCAV2026.05	1	3	0