Jailbreak Attack on HADES Self-harm
5.33ASRHADES
Evaluation Results
| Method | Links | |
|---|---|---|
| HADESModel=GPT-4o, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 5.33 | |
| HADESModel=InternVL 2.5, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 13.33 | |
| HADESModel=Qwen2.5-VL, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 20 | |
| SI-AModel=Qwen2.5-VL, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 32.67 | |
| SI-AModel=GPT-4o, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 32.67 | |
| SI-AModel=Grok-2-vision, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 33.33 | |
| SI-AModel=InternVL 2.5, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 35.33 | |
| HADESModel=Grok-2-vision, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 38 | |
| VisCRAModel=InternVL 2.5, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 44.67 | |
| VisCRAModel=GPT-4o, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 53.33 | |
| VisCRAModel=Qwen2.5-VL, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 68.67 | |
| GAMBITModel=GPT-4o, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 88 | |
| GAMBITModel=InternVL 2.5, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 90.67 | |
| GAMBITModel=Grok-2-vision, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 92.67 | |
| GAMBITModel=Qwen2.5-VL, Evaluation Protocol=Pass@5, Evaluator=Llama-Guard-32026.01 | 94.67 |