List your company

Jailbreak Defense on LLaVA v1.5

3.18ASRNullSteer
1.052815.411429.7744.1286Mar 23, 2026

Evaluation results

MethodLinks
NullSteerepsilon=16/2552026.033.18
ASTRAepsilon=16/2552026.034.55
NullSteerepsilon=32/2552026.038.75
NullSteerepsilon=unconstrained2026.0310.25
ASTRAepsilon=32/2552026.0310.91
NullSteerepsilon=64/2552026.0311.48
ASTRAepsilon=unconstrained2026.0312.43
Benign imageepsilon=16/2552026.0313.64
Benign imageepsilon=32/2552026.0313.64
Benign imageepsilon=64/2552026.0313.64
Benign imageepsilon=unconstrained2026.0313.64
ASTRAepsilon=64/2552026.0313.64
ECSOepsilon=64/2552026.0314.55
JailGuardepsilon=unconstrained2026.0317.27
Jailbreak Templatesepsilon=32/2552026.0317.27
Refusal Pairsepsilon=unconstrained2026.0319.09
ECSOepsilon=unconstrained2026.0320
Refusal Pairsepsilon=64/2552026.0320
Jailbreak Templatesepsilon=64/2552026.0320
JailGuardepsilon=32/2552026.0321.82
ECSOepsilon=32/2552026.0321.82
Self-reminderepsilon=64/2552026.0322.73
Self-reminderepsilon=unconstrained2026.0322.73
JailGuardepsilon=16/2552026.0323.64
Refusal Pairsepsilon=16/2552026.0323.64
Jailbreak Templatesepsilon=16/2552026.0323.64
ECSOepsilon=16/2552026.0324.55
Refusal Pairsepsilon=32/2552026.0325.45
Self-reminderepsilon=16/2552026.0328.18
Jailbreak Templatesepsilon=unconstrained2026.0329.09
Self-reminderepsilon=32/2552026.0330
JailGuardepsilon=64/2552026.0330
w/o defenseepsilon=16/2552026.0351.82
w/o defenseepsilon=unconstrained2026.0353.64
w/o defenseepsilon=64/2552026.0355.45
w/o defenseepsilon=32/2552026.0356.36