Malicious Action Detection on 100-seed stress orig profile (test)
100AMRPrior exploration baseline
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Prior exploration baselineProfile=orig2026.05 | 100 | 0 | 50 | |
| ToolShield (grouped replay)Profile=orig2026.05 | 10.87 | 0 | 94.57 | |
| VISTA-Guard precursor (grouped risk model)Profile=orig2026.05 | 4.35 | 1.85 | 96.9 | |
| MCPMark directProfile=orig2026.05 | 2.17 | 46.3 | 64.47 |