Autonomous Penetration Testing on Game of Active Directory (GOAD)
37.75Accounts Compromised per HourClaude-4.7-Opus
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Claude-4.7-OpusEvaluation Harness=cochise, Protocol=5 runs per model2026.05 | 37.75 | 21.35 | 0.74 | |
| Gemini-3-FlashEvaluation Harness=cochise, Protocol=5 runs per model2026.05 | 15.77 | 1.75 | 0.11 |