Jailbreak on Malicious-Educator
80Attack Success Rate (ASR)H-CoT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| H-CoTTarget Model=Qwen3-235B, Mode=transfer2026.05 | 80 | — | |
| H-CoTTarget Model=GPT-4o, Mode=transfer2026.05 | 84 | — | |
| H-CoTTarget Model=o3-mini2026.05 | 90 | 70.8 | |
| H-CoTTarget Model=Gemini-2.5, Mode=transfer2026.05 | 90 | — | |
| H-CoTTarget Model=DeepSeek-v3.1, Mode=transfer2026.05 | 90 | — | |
| AE-CoTTarget Model=Qwen3-235B, Mode=transfer2026.05 | 90 | — | |
| AE-CoTTarget Model=o3-mini2026.05 | 94 | 72 | |
| H-CoTTarget Model=Gemini-2.52026.05 | 96 | 80 | |
| H-CoTTarget Model=o1-mini2026.05 | 98 | 80 | |
| H-CoTTarget Model=GPT-3.5-turbo, Mode=transfer2026.05 | 98 | — | |
| AE-CoTTarget Model=GPT-4o, Mode=transfer2026.05 | 98 | — | |
| AE-CoTTarget Model=o1-mini2026.05 | 100 | 82 | |
| H-CoTTarget Model=DeepSeek-R12026.05 | 100 | 90 | |
| AE-CoTTarget Model=DeepSeek-R12026.05 | 100 | 93.6 | |
| AE-CoTTarget Model=Gemini-2.52026.05 | 100 | 84.6 | |
| AE-CoTTarget Model=Gemini-2.5, Mode=transfer2026.05 | 100 | — | |
| AE-CoTTarget Model=GPT-3.5-turbo, Mode=transfer2026.05 | 100 | — | |
| AE-CoTTarget Model=DeepSeek-v3.1, Mode=transfer2026.05 | 100 | — |