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

Jailbreak Detection on JailbreaksOverTime (test)

38.06Throughput (items/s)

MLJailDe

Updated 1mo ago

Evaluation Results

Method	Links
MLJailDe 2026.04		38.06	-	-	-
DeBERTa-ft 2026.04		37.71	-	-	-
JBShield 2026.04		32.05	-	-	-
PromptGuard 2026.04		6.629	-	-	-
GPT-4o-p 2026.04		0.8	-	-	-
GPT-4.1-p 2026.04		0.778	-	-	-
Qwen2.5-ft 2026.04		0.651	-	-	-
Llama-3.1-ft 2026.04		0.562	-	-	-
Llama-2-ft 2026.04		0.456	-	-	-
Moderation 2026.04		0.296	-	-	-
Qwen3-ft 2026.04		0.255	-	-	-
SelfReminder 2026.04		0.225	-	-	-
Claude-4.5-p 2026.04		0.168	-	-	-
SelfDefend 2026.04		0.158	-	-	-
GPT-5-p 2026.04		0.07	-	-	-
GPT-4o-p 2026.04		-	80.2	94.5	86.8
GPT-4.1-p 2026.04		-	88.7	92.4	90.5
GPT-5-p 2026.04		-	91.6	99.7	95.5
Claude-4.5-p 2026.04		-	85.5	100	92.2
Llama-2-ft 2026.04		-	56.8	13.9	22.4
Llama-3.1-ft 2026.04		-	81.5	40	53.7
Qwen2.5-ft 2026.04		-	74.2	34.8	47.4
Qwen3-ft 2026.04		-	72.1	36.1	48.1
SelfReminder 2026.04		-	97.3	75.8	85.2
SelfDefend 2026.04		-	57.7	64.9	61.1
DeBERTa-ft 2026.04		-	96.9	47.3	63.5
Moderation 2026.04		-	95	28.5	43.8
PromptGuard 2026.04		-	95.1	52.7	67.8
JBShield 2026.04		-	45.2	57.6	50.7
MLJailDe 2026.04		-	99.7	97.3	98.5