Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Prompt Injection Defense on Jailbreak MI-FGSM
Loading...
4
Attack Success Rate
GUARD
0.32
25.16
50
74.84
May 24, 2026
Attack Success Rate
Updated 8d ago
Evaluation Results
Method
Method
Links
Attack Success Rate
GUARD
Perturbation limit (ε)=16
2026.05
4
Gradient-guided Token Suppression
Perturbation limit (ε)=16
2026.05
4
Gradient-guided Token Suppression
Perturbation limit (ε)=32
2026.05
5
CIDER
Perturbation limit (ε)=16
2026.05
6
GUARD
Perturbation limit (ε)=32
2026.05
7
CIDER
Perturbation limit (ε)=32
2026.05
16
ECSO
Perturbation limit (ε)=16
2026.05
17
ECSO
Perturbation limit (ε)=32
2026.05
18
ASTRA
Perturbation limit (ε)=32
2026.05
33
ASTRA
Perturbation limit (ε)=16
2026.05
42
W/O DEFENSE
Perturbation limit (ε)=32
2026.05
95
W/O DEFENSE
Perturbation limit (ε)=16
2026.05
96
Feedback
Search any
task
Search any
task