Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

attack prompts

Benchmarks

Task NameDataset NameSOTA ResultTrend
Reasoning length evaluation20 attack prompts
Avg Length7,745
48
System Prompt Extractionattack prompts Out-of-domain
ASR (TA)78
5
System Prompt ExtractionAttack Prompts In-domain
Attack Success Rate (TA)78
5
JailbreakingAttack Prompts Out-of-domain
ASR (TA)67.7
5
JailbreakingAttack Prompts In-domain
ASR (TA)76
5
Prompt InjectionAttack prompts Out-of-domain
ASR (TA)47
5
Prompt InjectionIn-domain attack prompts
ASR (TA)53
5
BackdoorAttack prompts Out-of-domain
ASR (TA)99
5
Backdoorattack prompts In-domain
ASR (TA)100
5
Showing 9 of 9 rows