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

ASB

Benchmarks

Task NameDataset NameSOTA ResultTrend
Guarded Agent EvaluationASB latest (IPI)
ASR5.5
14
Guarded Agent EvaluationASB latest (DPI)
ASR95.25
14
Agent SecurityASB (Agent Security Benchmark)
No Attack UA90
8
Agent Security EvaluationASB (Agent Security Benchmark)
ASR-d (ASB)7
8
Agent SecurityASB n=2000
ASR0
5
Secure LLM Agent Task CompletionASB
Benign Utility78.75
4
Prompt Injection DefenseASB
ASR9.1
3
Agent Utility EvaluationASB 5 direct prompt-injection styles
Utility (%)88.9
2
Prompt Injection PreventionASB 5 direct prompt-injection styles
ASR0
2
Showing 9 of 9 rows