Our new X account is live! Follow @wizwand_team for updates

Jailbreak Robustness on JBB-Behaviors

0.3ASR (PAIR, Guardrail Model)

RA (ours)

Updated 4d ago

Evaluation Results

Method	Links
RA (ours) 2025.10		0.3	10	45	60	29.3	67.3
MOSA 2025.10		1	27.7	65	56	54.3	74
RA (ours) 2025.10		1.3	21	45	56.7	38.7	58.3
MOSA 2025.10		2	27.3	67.7	56.7	57	74.3
RA w/o inter (ablation) 2025.10		2.3	36	68	62.5	53.7	72.3
DPO 2025.10		2.7	33.3	69	57.3	60.7	84.7
DPO 2025.10		3	37.7	70.3	61.3	66.7	83.7
SFT 2025.10		4	39.3	68.7	56	70	85.3
RA w/o inter (ablation) 2025.10		4	25.7	68.3	60.3	52.3	68.7
Original 2025.10		4.7	44.7	74.3	63	68.3	91.7
Original 2025.10		5	41.7	77.3	65	68	89.7
SFT 2025.10		5.3	46.3	71	54.7	67.7	80.7
RA (ours) 2025.10		13	42.3	50	65	65.3	90.7
MOSA 2025.10		17.7	46.3	68.3	65.7	55.3	83.7
RA w/o inter (ablation) 2025.10		32	52.4	60	85	66.3	86.7
DPO 2025.10		37.5	60	79	65.7	61.3	85.7
SFT 2025.10		47.7	70.3	86.5	78.5	57	71.3
Original 2025.10		55	81	81.7	85.7	64	88.3