Jailbreak Success Evaluation on Merged Jailbreak Evaluation Set KIMI-verified 1.0 (test)
6,756Malware TP CountHomotopy-Inspired Prompt Obfuscation
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Homotopy-Inspired Prompt ObfuscationTarget LLM=KIMI2026.01 | 6,756 | 1,969 | 77.43 | 77.43 | 22.57 | |
| Homotopy-Inspired Prompt ObfuscationTarget LLM=DeepSeek2026.01 | 403 | 97 | 80.6 | 80.6 | 19.4 | |
| Homotopy-Inspired Prompt ObfuscationTarget LLM=Llama2026.01 | 311 | 189 | 62.2 | 62.2 | 37.8 |