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

JBShield

Benchmarks

Task NameDataset NameSOTA ResultTrend
Jailbreak DetectionJBShield evaluation suite GCG attack on Llama-3-8B
Detection Accuracy100
4
Jailbreak DetectionJBShield evaluation suite Base64 attack on Llama-3-8B
Detection Accuracy100
2
Jailbreak DetectionJBShield evaluation suite Zulu attack on Llama-3-8B
Accuracy100
2
Jailbreak DetectionJBShield Puzzler attack on Llama-3-8B
Detection Accuracy100
2
Jailbreak DetectionJBShield evaluation suite DrAttack attack on Llama-3-8B
Detection Accuracy100
2
Jailbreak DetectionJBShield PAIR attack on Llama-3-8B
Detection Accuracy77
2
Jailbreak DetectionJBShield AutoDAN attack on Llama-3-8B
Detection Accuracy97
2
Jailbreak DetectionJBShield evaluation suite IJP attack on Llama-3-8B
Detection Accuracy96
2
Jailbreak Attack Success RateJBShield (test)
Attack Success Rate (ASR)95
1
Showing 9 of 9 rows