| ImpossibleBench | | Average Precision100 | | 40 | 3mo ago |
| TRACE | | Average Precision90.1 | | 28 | 3mo ago |
| BigCodeBench Sabotage (reasoning LLM attacker) | Extract-and-Evaluate | log-AUROC0.87 | | 8 | 4mo ago |
| BigCodeBench-Sabotage traditional LLM attacker | Extract-and-Evaluate | log-AUROC0.84 | | 8 | 4mo ago |
| MLE-Sabotage | Action-only | log-AUROC0.87 | | 8 | 4mo ago |
| SHADE-Arena | CoT+action | log-AUROC78 | | 8 | 4mo ago |
| SHADE-Arena Total | Sequential | Accuracy70.6 | | 3 | 1mo ago |
| SHADE-Arena Charity Allocation | Full-trajectory | Accuracy91.4 | | 3 | 1mo ago |
| SHADE-Arena Performance Reviews | TRACE | Accuracy67.3 | | 3 | 1mo ago |
| SHADE-Arena Expenditure Accounting | Full-trajectory | Accuracy87.9 | | 3 | 1mo ago |
| SHADE-Arena API Key Calendar | Full-trajectory | Accuracy88.5 | | 3 | 1mo ago |
| SHADE-Arena Jira Tickets | Sequential | Accuracy72.5 | | 3 | 1mo ago |
| SHADE-Arena Election Tampering | TRACE | Accuracy44.4 | | 3 | 1mo ago |
| SHADE-Arena Spam Filter | Sequential | Accuracy69.5 | | 3 | 1mo ago |
| SHADE-Arena American Airlines | Sequential | Accuracy83 | | 3 | 1mo ago |
| SHADE-Arena Bank Transfer | Sequential | Accuracy85.7 | | 3 | 1mo ago |
| SHADE-Arena SSN Task | Full-trajectory | Accuracy96.3 | | 3 | 1mo ago |
| DSTS (Distributed Sub-Threshold Sabotage) | Monitor A | Max Power (A)0.048 | | 1 | 1mo ago |