List your company

Jailbreak Defense on MiniGPT-4

7.32Attack Success Rate (ASR)NullSteer
5.467217.973630.4842.9864Mar 23, 2026

Evaluation results

MethodLinks
NullSteerepsilon=unconstrained2026.037.32
NullSteerepsilon=16/2552026.037.62
ASTRAepsilon=16/2552026.039.09
ASTRAepsilon=unconstrained2026.039.09
NullSteerepsilon=32/2552026.0310.72
ASTRAepsilon=32/2552026.0313.18
NullSteerepsilon=64/2552026.0313.58
ASTRAepsilon=64/2552026.0315.46
Refusal Pairsepsilon=unconstrained2026.0316.36
Refusal Pairsepsilon=64/2552026.0317.27
Refusal Pairsepsilon=16/2552026.0320
Refusal Pairsepsilon=32/2552026.0322.73
Benign imageepsilon=16/2552026.0324.55
Benign imageepsilon=32/2552026.0324.55
Benign imageepsilon=64/2552026.0324.55
Benign imageepsilon=unconstrained2026.0324.55
JailGuardepsilon=64/2552026.0327.27
JailGuardepsilon=unconstrained2026.0328.18
ECSOepsilon=64/2552026.0329.09
JailGuardepsilon=16/2552026.0330
JailGuardepsilon=32/2552026.0332.73
Jailbreak Templatesepsilon=16/2552026.0333.64
Self-reminderepsilon=16/2552026.0335.45
Self-reminderepsilon=32/2552026.0336.36
ECSOepsilon=unconstrained2026.0337.27
Jailbreak Templatesepsilon=32/2552026.0338.15
Jailbreak Templatesepsilon=64/2552026.0338.18
ECSOepsilon=16/2552026.0340.91
Self-reminderepsilon=64/2552026.0341.82
ECSOepsilon=32/2552026.0342.73
Jailbreak Templatesepsilon=unconstrained2026.0342.73
Self-reminderepsilon=unconstrained2026.0343.64
w/o defenseepsilon=16/2552026.0344.55
w/o defenseepsilon=32/2552026.0347.27
w/o defenseepsilon=64/2552026.0349.09
w/o defenseepsilon=unconstrained2026.0353.64