SIA-Bench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
SIA-Bench Held-out indirect prompt injection
43.9HR-AUPRC
7
SIA-Bench Locked in-distribution (test)
20.4HR-AUPRC
7
SIA-Bench SkillJect-derived threat slice (ood)
40.5HR-AUPRC
2
SIA-Bench SkillJect-derived threat slice (test)
32.7HR-AUPRC
2
SIA-Bench IPIGuard replay (ood)
40.5HR-AUPRC
2
SIA-Bench IPIGuard replay (test)
32.7HR-AUPRC
2