ResearchTasksPrompt injection detectionFollowBenchmarksDataset NameSOTA MethodSortMost resultsRecently updatedMost papersApplyDataset NameSOTA methodMetricTrendResultsLast UpdatedIndirect-Harmful ShiftPrompt-Guard-2 (86M)21.7FNR2Jun 23, 202616 public datasets (5-fold StratifiedKFold)Gate98.7Micro F1 Score2Jun 4, 2026Hard-negative injection OODTF–IDF only80.82Macro F1 Score2May 27, 2026Primary HarmBench + Alpaca (OOD)DeBERTa-base75.28Macro-F12May 27, 2026Camouflage Injection DatasetLlama 3.1 8B18.9ASR (static)2May 22, 2026Page 3 of 3PreviousNext