Safety Judge Evaluation on High-Confidence (HC) (test)
99.9F1 ScoreClaude
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Claude2026.06 | 99.9 | — | 100 | 99.7 | 99.7 | 99.9 | |
| SorryBench2026.06 | 98.2 | — | 100 | 96.5 | 96.6 | 98.2 | |
| AILuminate2026.06 | 96.8 | — | 100 | 93.7 | 94 | 96.9 | |
| StrongReject2026.06 | 96.4 | — | 100 | 93.1 | 96.3 | 99.7 | |
| Ettin-400MSubset=HQ2026.06 | 90 | 88.7 | 90.4 | 89.6 | 92.2 | 98.4 | |
| Ettin-150MSubset=HQ2026.06 | 85.3 | 83.4 | 90 | 81 | 88.3 | 97.8 | |
| Ettin-400MSubset=NF2026.06 | 85.1 | 83.2 | 98.4 | 75 | 96 | 99.3 | |
| JBB2026.06 | 83.5 | — | 100 | 71.6 | 72.8 | 85.8 | |
| Ettin-150MSubset=NF2026.06 | 80 | 78.1 | 95.6 | 68.8 | 93 | 98.8 | |
| LG32026.06 | 75.2 | — | 100 | 60.3 | 61.9 | 80.1 | |
| LG42026.06 | 49.9 | — | 100 | 33.2 | 36 | 66.6 | |
| ShieldGemma-9B2026.06 | 16.3 | 15.1 | 9.8 | 48.6 | 8.9 | 73.7 | |
| ShieldGemma-2B2026.06 | 15.2 | 13.5 | 10.8 | 25.6 | 8.4 | 67.5 |