Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Prefill Attack Robustness on AdvBench 50 OOD harmful prompts-prefill pairs 2025
Loading...
80
PAR
ACT
-3.2
18.4
40
61.6
Jun 4, 2026
PAR
Updated 1mo ago
Evaluation Results
Method
Method
Links
PAR
ACT
Backbone=Gemma-3-27B-it
2026.06
80
Baseline (no defense)
Backbone=Llama-3.1-8B-...
2026.06
52
ACT
Backbone=Llama-3.1-8B-...
2026.06
50
AttCT
Backbone=Llama-3.1-8B-...
2026.06
46
Baseline (no defense)
Backbone=Gemma-3-27B-it
2026.06
36
AttCT
Backbone=Gemma-3-27B-it
2026.06
36
BCT
Backbone=Llama-3.1-8B-...
2026.06
0
MLPCT+BCT
Backbone=Llama-3.1-8B-...
2026.06
0
BCT
Backbone=Gemma-3-27B-it
2026.06
0
MLPCT+BCT
Backbone=Gemma-3-27B-it
2026.06
0
Feedback
Search any
task
Search any
task