Safety Classification on Safety OOD (holdout)
76.7F1 ScoreAILuminate
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AILuminate2026.06 | 76.7 | 74.3 | 79.2 | 20.8 | 5.1 | |
| SorryBench2026.06 | 70.6 | 58.5 | 88.8 | 11.2 | 7.3 | |
| JBB2026.06 | 61.5 | 78.1 | 50.7 | 49.3 | 3.1 | |
| Ettin-150M2026.06 | 54.7 | 51.9 | 57.8 | 42.2 | 4.8 | |
| Ettin-400M2026.06 | 44.9 | 30.4 | 85.4 | 14.6 | 13.4 | |
| LG42026.06 | 38.5 | 44 | 34.2 | 65.8 | 3.5 | |
| Claude2026.06 | 31.9 | 19 | 99.4 | 0.6 | 25.1 | |
| LG32026.06 | 30.9 | 23.6 | 44.7 | 55.3 | 9.1 | |
| StrongReject2026.06 | 28.9 | 17.2 | 90 | 10 | 25.1 | |
| ShieldGemma-9B2026.06 | 15.4 | 9 | 52.4 | 47.6 | 27.9 | |
| PrefixMatch (baseline)2026.06 | 11.6 | 6.2 | 96.3 | 3.7 | 74.9 | |
| ShieldGemma-2B2026.06 | 5.3 | 6.3 | 4.6 | 95.4 | 3.5 |