| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| Taylor | TELLME | Accuracy99 | 63 | 1mo ago | |
| WorkBench (n=326) | Average IR52 | 30 | 22d ago | ||
| Agent-SafetyBench n=35 (aligned traces) | DirectPrompt | Avg. IR31.4 | 30 | 22d ago | |
| Workbench n=62 (general) | ProvenanceGuard | Error Rate (ER)0.3 | 30 | 22d ago | |
| Agent-SafetyBench underspecified queries n=148 | Error Rate (ER)0.8 | 30 | 22d ago | ||
| MoralChain (test) | Linear Probe | Accuracy89.4 | 6 | 2mo ago | |
| MoralChain (train) | Linear Probe | Accuracy97.2 | 6 | 2mo ago | |
| All datasets 9 5 in-dist + 4 OOD (pooled) | 18-indicator ensemble | AUROC93 | 4 | 1mo ago | |
| Sycophancy Bloom behaviors (in-distribution) | RepE baseline | AUROC93.5 | 4 | 1mo ago | |
| Strategic deception Bloom behaviors (in-distribution) | 18-indicator ensemble | AUROC0.817 | 4 | 1mo ago | |
| Self-preservation Bloom behaviors (in-distribution) | 18-indicator ensemble | AUROC95.6 | 4 | 1mo ago | |
| Strategic sandbagging Bloom behaviors (in-distribution) | 18-indicator ensemble | AUROC0.985 | 4 | 1mo ago | |
| Covert code sabotage Bloom behaviors (in-distribution) | 18-indicator ensemble | AUROC0.959 | 4 | 1mo ago |
