LLM Jailbreaking on Gemma-DA

1SRFRepE

Evaluation results

Method
RepE2026.05	1	0	1
Angular2026.05	23	18	26
RD-C2026.05	37	50	55
SCAV2026.05	38	47	64
RD-A2026.05	57	57	76
Adaptive Probe-based Steering2026.05	69	73	88