Jailbreak Detection on OKT
1Correlation ScoreLlama-Prompt-Guard-2
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama-Prompt-Guard-2Parameters=86M2026.02 | 1 | |
| AISABackbone=GPT-OSS-20b-I2026.02 | 0.9933 | |
| AISABackbone=Llama3.1-8b-I2026.02 | 0.99 | |
| AISABackbone=Llama2-13b-I2026.02 | 0.99 | |
| AISABackbone=Mistral-7b-I2026.02 | 0.98 | |
| AISABackbone=Qwen3-8b-I2026.02 | 0.98 | |
| GradSafe2026.02 | 0.9733 | |
| gpt-4.1-miniVersion=2025-04-142026.02 | 0.97 | |
| gpt-5-miniVersion=2025-08-072026.02 | 0.9667 | |
| gpt-4o-miniVersion=2024-07-182026.02 | 0.9467 | |
| Jailbreak-Classifier2026.02 | 0.88 | |
| NemoGuard-JailbreakDetect2026.02 | 0.8633 | |
| SPDetector2026.02 | 0.6667 |