Safety Evaluation on HVD-Implicit
1.77HCRateGPT-4o
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-4oAttack=None2026.06 | 1.77 | 74.49 | |
| Claude-3.7-SonnetAttack=None2026.06 | 4.42 | 51.17 | |
| DeepSeek-V3Attack=None2026.06 | 20.35 | 87.11 | |
| GHOSTWRITERTarget Model=GPT-4o, Attacker LM=GPT-4o-mini2026.06 | 53.1 | 83.68 | |
| GHOSTWRITERTarget Model=Claude-3.7-Sonnet, Attacker LM=GPT-4o-mini2026.06 | 97.35 | 92.87 | |
| GHOSTWRITERTarget Model=DeepSeek-V3, Attacker LM=GPT-4o-mini2026.06 | 147.79 | 92.73 | |
| GHOSTWRITERTarget Model=GPT-4o, Attacker LM=Qwen-2.5-7B2026.06 | 175.22 | 84.38 | |
| GHOSTWRITERTarget Model=Claude-3.7-Sonnet, Attacker LM=Qwen-2.5-7B2026.06 | 209.73 | 94.11 | |
| GHOSTWRITERTarget Model=DeepSeek-V3, Attacker LM=Qwen-2.5-7B2026.06 | 293.81 | 92.33 |