List your company

Robustness against priming vulnerability on JBB-Behaviors (test)

0ASR (Guardrail Model)RA (ours)
-349.22,007.94,3656,722.1Oct 1, 2025

Evaluation results

MethodLinks
RA (ours)Backbone=LLaDA, Attack=No Attack2025.100130
RA (ours)Backbone=LLaDA, Attack=Anchoring, t_inter=12025.1000
RA (ours)Backbone=LLaDA, Attack=Anchoring, t_inter=42025.100170
RA (ours)Backbone=LLaDA1.5, Attack=No Attack2025.100430
RA (ours)Backbone=LLaDA, Attack=Anchoring, t_inter=82025.1070100
OriginalBackbone=LLaDA1.5, Attack=No Attack2025.1070500
RA (ours)Backbone=LLaDA, Attack=PAD2025.10100230
RA (ours)Backbone=LLaDA, Attack=Anchoring, t_inter=162025.10200670
RA (ours)Backbone=MMaDA, Attack=No Attack2025.10400570
RA (ours)Backbone=LLaDA, Attack=First-Step GCG2025.105301,230
RA (ours)Backbone=LLaDA, Attack=DiJA2025.101,1701,930
RA (ours)Backbone=LLaDA1.5, Attack=Anchoring, t_inter=322025.102,7301,770
RA (ours)Backbone=LLaDA, Attack=Anchoring, t_inter=322025.102,8301,430
RA (ours)Backbone=MMaDA, Attack=PAD2025.102,8302,330
MOSABackbone=MMaDA, Attack=No Attack2025.102,8702,600
RA (ours)Backbone=MMaDA, Attack=First-Step GCG2025.104,3004,830
RA (ours)Backbone=MMaDA, Attack=DiJA2025.105,8003,870
OriginalBackbone=MMaDA, Attack=No Attack2025.107,4006,100
RA (ours)Backbone=MMaDA, Attack=Anchoring, t_inter=322025.107,6303,830
RA w/o inter (ablation)Backbone=LLaDA, Attack=Anchoring, t_inter=322025.108,7306,070