Agent Memory Attack Defense on ReAct-QA
6Attack Success RateSENTINEL
Evaluation Results
| Method | Links | |
|---|---|---|
| SENTINELModel=GPT-4o-mini2026.07 | 6 | |
| A-MemG. (heur.)Model=GPT-4o-mini2026.07 | 46 | |
| NoDef.Model=GPT-4o-mini2026.07 | 52 | |
| KeyFilt.Model=GPT-4o-mini2026.07 | 52 |
| Method | Links | |
|---|---|---|
| SENTINELModel=GPT-4o-mini2026.07 | 6 | |
| A-MemG. (heur.)Model=GPT-4o-mini2026.07 | 46 | |
| NoDef.Model=GPT-4o-mini2026.07 | 52 | |
| KeyFilt.Model=GPT-4o-mini2026.07 | 52 |