Jailbreak Attack on StrategyQA (ASR, HS)
3Attack Success Rate (ASR)H-COT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| H-COTVictim LRM=Qwen2.5-Max2026.04 | 3 | 2.33 | |
| H-COTVictim LRM=OpenAI o4-mini2026.04 | 4 | 2 | |
| Cognitive OverloadVictim LRM=Qwen2.5-Max2026.04 | 6 | 2 | |
| Cognitive OverloadVictim LRM=OpenAI o4-mini2026.04 | 20 | 2.05 | |
| H-COTVictim LRM=DeepSeek R12026.04 | 22 | 2.82 | |
| Cognitive OverloadVictim LRM=DeepSeek R12026.04 | 28 | 2.5 | |
| PRJAVictim LRM=OpenAI o4-mini2026.04 | 78 | 2.76 | |
| PRJAVictim LRM=Qwen2.5-Max2026.04 | 84 | 2.93 | |
| PRJAVictim LRM=DeepSeek R12026.04 | 86 | 2.97 |