Prompt Injection Defense on Indirect Prompt Injection Tail 1.0 benchmark leaderboard · SOTA2 Research

Prompt Injection Defense on Indirect Prompt Injection Tail 1.0

0.11ASR NaiveExtraction removal method

Evaluation Results

Method
Extraction removal method2025.02	0.11	1	0.11	1.77	3.44
StruQ2025.02	0.11	4.77	4.33	86.22	79.33
Extraction removal method2025.02	0.11	1.22	0.11	2.11	3.77
Extraction removal method2025.02	0.22	0.22	0.33	2.44	4.22
StruQ2025.02	0.66	2.55	10.55	75.33	72
Segmentation removal method2025.02	0.77	0.11	0.77	16.88	0.11
Segmentation removal method2025.02	3.11	0.11	3.11	14.44	0.11
StruQ2025.02	4.22	1.22	6.44	94.55	84.66
Segmentation removal method2025.02	5.88	0.11	5.88	14.88	0.11
Sandwich2025.02	10.88	16.11	25.77	12.11	14.77
Sandwich2025.02	13	17.88	22.33	18.33	24.44
Sandwich2025.02	24	29.66	27.11	54.11	57.22
Instructional2025.02	39.11	51.33	64.33	64.88	65
Instructional2025.02	41.22	55.33	59.88	95.66	88
None2025.02	46.88	69.22	69.22	75.77	83
None2025.02	47.44	63.33	65.66	96.44	93.66
Instructional2025.02	52.77	58.55	63.66	73.66	80
None2025.02	63.77	76.11	75.44	85.44	86.11