Formal verification of safety classifiers on Toxigen
0.9tau* ScoreGaussian Mixture Models
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gaussian Mixture ModelsVerification type=GMM, Model=GPT-2, K=3, Cov=FULL2026.05 | 0.9 | 0.98 | |
| Gaussian Mixture ModelsVerification type=GMM, Model=Llama, K=3, Cov=DIAG2026.05 | 0.8 | 0.8 | |
| Gaussian Mixture ModelsVerification type=GMM, Model=BERT, K=1, Cov=FULL2026.05 | 0.55 | 1 |