Jailbreak Safety Evaluation on English dataset Multi-Image
14StrongREJECT (Perturbed)GPT-4o
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-4o2025.04 | 14 | 13 | 20 | |
| Claude 3.5 Sonnet2025.04 | 10 | 9 | 0 | |
| Gemini 1.5 Flash2025.04 | 8 | 8 | 32 | |
| Gemini 1.5 Pro2025.04 | 8 | 7 | 8 | |
| Claude 3 Haiku2025.04 | 7 | 8 | 9 | |
| GPT-4o Mini2025.04 | 1 | 1 | 20 |