Tool Selection Hijacking on LLM agent tool-selection tasks
69.8Attack Success Rate (ASR)MemMorph
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MemMorphDefense=Distil Classifier2026.05 | 69.8 | 84.5 | |
| MemMorphDefense=PPL Filter2026.05 | 65.8 | 82.1 | |
| MemMorphDefense=Memory Auditor2026.05 | 54.9 | 72.3 | |
| PoisonedRAGDefense=PPL Filter2026.05 | 38.6 | 52.1 | |
| PoisonedRAGDefense=Distil Classifier2026.05 | 29.4 | 44.8 | |
| PoisonedRAGDefense=Memory Auditor2026.05 | 22.7 | 36.3 | |
| GCGDefense=Distil Classifier2026.05 | 21.3 | 31.5 | |
| GCGDefense=Memory Auditor2026.05 | 16.7 | 24.1 | |
| GCGDefense=PPL Filter2026.05 | 12.4 | 18.7 |