Indirect Prompt Injection Detection on TriviaQA Inj benchmark leaderboard · SOTA2 Research

Indirect Prompt Injection Detection on TriviaQA Inj

0.22FPR (None)Prompt-Guard

Evaluation Results

Method	Links
Prompt-Guard2025.02		0.22	77.77	85.66	77.77	39.55	73.66
Prompt-Guard2025.02		0.22	88.66	94.33	89.33	86	92.88
Prompt-Guard2025.02		0.22	69	78.55	69	70.88	78.33
Protect-AI-detector2025.02		0.66	0.33	43.77	0.33	2.88	63.88
Protect-AI-detector2025.02		0.66	0.33	28.55	0.33	0.44	29.22
Protect-AI-detector2025.02		0.66	0.22	23.22	0.22	0.44	27.88
Llama-Guard3-8B2025.02		1	2.88	5	2.88	2.11	4
Llama-Guard3-8B2025.02		1	1	2.22	1.22	2	2.44
Llama-Guard3-8B2025.02		1	6.77	29.33	9	6.66	16.44
Trained-Qwen2-1.5B2025.02		11.11	90.22	97	93.11	94.77	98
Trained-Qwen2-1.5B2025.02		11.11	94.11	99	97	97.22	99.55
Trained-Qwen2-1.5B2025.02		11.11	99.11	100	99.33	99.66	100
Trained-DeBERTa2025.02		12.44	98.11	100	98.11	92.22	99.88
Trained-DeBERTa2025.02		12.44	99.55	100	99.66	99.88	100
Trained-DeBERTa2025.02		12.44	99.77	100	99.77	99.88	100
Trained-Qwen2-0.5B2025.02		27.33	96	97.55	94.66	95.44	97.33
Trained-Qwen2-0.5B2025.02		27.33	95	98.22	97.66	97.44	98.88
Trained-Qwen2-0.5B2025.02		27.33	97.55	99.66	97.66	98.55	99.55
Qwen2-7B-Instruct2025.02		34.33	32.22	44.11	35.33	39.55	61.11
Qwen2-7B-Instruct2025.02		34.33	47.55	66.66	58	66.77	64
Qwen2-7B-Instruct2025.02		34.33	58.55	60.66	55.22	47.22	63.33
Llama3-8B-Instruct2025.02		76.11	96.88	97.77	96.11	100	100
Llama3-8B-Instruct2025.02		76.11	84	92.44	86.33	98.55	97.11
Llama3-8B-Instruct2025.02		76.11	89.22	91.88	84.33	97.44	92.88