Watermark Evasion on LLM Watermarking Algorithms KGW, Unigram, UPV, EWD, DIP, SIR, EXP
99.8KGW Evasion ScoreBIRA
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| BIRABackbone=Llama-3.1-8B2025.09 | 99.8 | 99.4 | 99.8 | 100 | 100 | 99.6 | 99.8 | |
| BIRABackbone=Llama-3.1-70B2025.09 | 99.4 | 99 | 99.6 | 99.8 | 99.6 | 98.8 | 98 | |
| BIRABackbone=GPT-4o-mini2025.09 | 99.4 | 100 | 100 | 99.8 | 99.8 | 99.4 | 98.2 | |
| SIRABackbone=Llama-3.1-8B2025.09 | 98.8 | 95 | 87.6 | 99.8 | 99.6 | 72.8 | 95.2 | |
| SIRABackbone=Llama-3.1-70B2025.09 | 98 | 87.6 | 85 | 99.2 | 99.6 | 60.6 | 88.6 | |
| SIRABackbone=GPT-4o-mini2025.09 | 98 | 85.2 | 84.8 | 97.2 | 99.6 | 57.6 | 94.8 | |
| DIPPER-2Configuration=22025.09 | 97.2 | 71.8 | 85.4 | 96.6 | 99.2 | 70.4 | 97.2 | |
| DIPPER-1Configuration=12025.09 | 93.8 | 61.2 | 80.6 | 92.8 | 99.4 | 55.6 | 90.8 | |
| VanillaBackbone=Llama-3.1-8B2025.09 | 88.8 | 73.4 | 73.4 | 92.6 | 99.8 | 54 | 80.6 | |
| VanillaBackbone=Llama-3.1-70B2025.09 | 87.4 | 67 | 65 | 89.4 | 98.8 | 42.8 | 70.4 | |
| VanillaBackbone=GPT-4o-mini2025.09 | 60.2 | 30.2 | 46.8 | 58.8 | 95.8 | 23.6 | 31.8 |