Jailbreak Detection on HB
100Correctness Rate (COR)gpt-4o-mini
Evaluation Results
| Method | Links | |
|---|---|---|
| gpt-4o-miniVersion=2024-07-182026.02 | 100 | |
| gpt-5-miniVersion=2025-08-072026.02 | 100 | |
| AISABackbone=Qwen3-8b-I2026.02 | 99.5 | |
| AISABackbone=Mistral-7b-I2026.02 | 99 | |
| AISABackbone=Llama2-13b-I2026.02 | 99 | |
| AISABackbone=Llama3.1-8b-I2026.02 | 98.5 | |
| gpt-4.1-miniVersion=2025-04-142026.02 | 98 | |
| Jailbreak-Classifier2026.02 | 98 | |
| AISABackbone=GPT-OSS-20b-I2026.02 | 97.5 | |
| GradSafe2026.02 | 95.5 | |
| NemoGuard-JailbreakDetect2026.02 | 87.5 | |
| SPDetector2026.02 | 68 | |
| Llama-Prompt-Guard-2Parameters=86M2026.02 | 20 |