Semantic Hijacking Attack Success Evaluation on Semantic Hijacking Evaluation Set
94.4ASRLlama-3.1-8B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Llama-3.1-8BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 94.4 | — | — | — | |
| Mistral-Small-3.2-24BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 91 | — | — | — | |
| Qwen-2.5-72BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 91 | — | — | — | |
| Qwen-2.5-7BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 89.4 | — | — | — | |
| Qwen-3-32BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 82.4 | — | — | — | |
| Qwen-2.5-7BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 79.6 | — | — | — | |
| DeepSeek-R1Role=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 76.4 | — | — | — | |
| Llama-3.1-8BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 75.8 | — | — | — | |
| Qwen-2.5-72BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 74.4 | — | — | — | |
| DeepSeek-R1Role=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 74.4 | — | — | — | |
| Mistral-NemoRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 72.2 | — | — | — | |
| Llama-3.1-70BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 71 | — | — | — | |
| Mistral-Small-3.2-24BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 66 | — | — | — | |
| Llama-3.1-8BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 65.4 | — | — | — | |
| Llama-3.1-70BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 65.2 | — | — | — | |
| GPT-4o-MiniRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 64.8 | — | — | — | |
| Qwen-2.5-7BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 62 | — | — | — | |
| Mistral-Small-3.2-24BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 60 | — | — | — | |
| Qwen-2.5-7BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 59.2 | — | — | — | |
| Mistral-Small-3.2-24BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 58.6 | — | — | — | |
| Qwen-3-32BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 58.4 | — | — | — | |
| Qwen-2.5-72BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 56.4 | — | — | — | |
| Mistral-NemoRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 51.2 | — | — | — | |
| GPT-4o-MiniRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 50.4 | — | — | — | |
| Qwen-2.5-72BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 49.6 | — | — | — | |
| Qwen-2.5-7BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 49 | — | — | — | |
| Llama-3.1-70BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 47.6 | — | — | — | |
| Mistral-Small-3.2-24BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 47.4 | — | — | — | |
| Llama-3.1-70BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 44.4 | — | — | — | |
| Qwen-2.5-7BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 43.8 | — | — | — | |
| Mistral-Small-3.2-24BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 43.4 | — | — | — | |
| Llama-3.1-8BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 42.2 | — | — | — | |
| Qwen-3-32BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 41.2 | — | — | — | |
| Mistral-NemoRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 40.8 | — | — | — | |
| Llama-3.1-8BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 40.6 | — | — | — | |
| Llama-3.1-8BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 38.2 | — | — | — | |
| Qwen-3.5-27BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 37.2 | — | — | — | |
| Qwen-2.5-7BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 33.8 | — | — | — | |
| Mistral-Small-3.2-24BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 33.6 | — | — | — | |
| GPT-4o-MiniRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 31.8 | — | — | — | |
| Qwen-3-32BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 31.8 | — | — | — | |
| Qwen-2.5-72BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 31 | — | — | — | |
| Llama-3.1-70BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 30.2 | — | — | — | |
| Qwen-2.5-72BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 29 | — | — | — | |
| Gemma-4-31BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 26.6 | — | — | — | |
| DeepSeek-R1Role=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 25.6 | — | — | — | |
| Qwen-2.5-72BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 23.4 | — | — | — | |
| Mistral-NemoRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 22.2 | — | — | — | |
| GPT-4o-MiniRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 22 | — | — | — | |
| Llama-3.1-70BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 21.6 | — | — | — | |
| DeepSeek-R1Role=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 21.6 | — | — | — | |
| Llama-3.1-8BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 20.6 | — | — | — | |
| Llama-3.1-70BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 20.2 | — | — | — | |
| DeepSeek-R1Role=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 20.2 | — | — | — | |
| DeepSeek-R1Role=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 19.4 | — | — | — | |
| Gemma-4-31BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 17 | — | — | — | |
| GPT-4o-MiniRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 15.6 | — | — | — | |
| GPT-4o-MiniRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 15.6 | — | — | — | |
| Qwen-3.5-9BRole=Manager, Worker Model=DeepSeek-R1, Configuration=Config A2026.05 | 14.6 | — | — | — | |
| Qwen-3-32BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 14.4 | — | — | — | |
| Qwen-3.5-27BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 14 | — | — | — | |
| Qwen-3-32BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 13.8 | — | — | — | |
| Mistral-NemoRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 13 | — | — | — | |
| Qwen-3.5-9BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 12.6 | — | — | — | |
| Gemma-4-31BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 12 | — | — | — | |
| Qwen-3.5-27BRole=Manager, Worker Model=Llama-3.1-70B, Configuration=Config A2026.05 | 11.6 | — | — | — | |
| Mistral-NemoRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 11.4 | — | — | — | |
| Qwen-3.5-9BRole=Manager, Worker Model=Qwen-2.5-7B, Configuration=Config A2026.05 | 11.2 | — | — | — | |
| Qwen-3-32BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 8.4 | — | — | — | |
| Qwen-3.5-27BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 6.4 | — | — | — | |
| Qwen-3.5-9BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 6.4 | — | — | — | |
| GPT-4o-MiniRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 6 | — | — | — | |
| Qwen-3.5-27BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 6 | — | — | — | |
| Qwen-3.5-9BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 5.8 | — | — | — | |
| Gemma-4-31BRole=Manager, Worker Model=Qwen-2.5-72B, Configuration=Config A2026.05 | 5.6 | — | — | — | |
| Gemma-4-31BRole=Manager, Worker Model=GPT-4o, Configuration=Config A2026.05 | 5 | — | — | — | |
| Mistral-NemoRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 4.4 | — | — | — | |
| DeepSeek-R1Role=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 3.8 | — | — | — | |
| Qwen-3.5-27BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 3 | — | — | — | |
| Gemma-4-31BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 2 | — | — | — | |
| Qwen-3.5-9BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 1.8 | — | — | — | |
| Qwen-3.5-27BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 1.6 | — | — | — | |
| Qwen-3.5-9BRole=Manager, Worker Model=Mistral-Small-3.2-24B, Configuration=Config A2026.05 | 1.4 | — | — | — | |
| Gemma-4-31BRole=Manager, Worker Model=Llama-3.1-8B, Configuration=Config A2026.05 | 0.8 | — | — | — | |
| DeepSeek-R1Role=Manager2026.05 | — | 34.49 | — | — | |
| DeepSeek-R1Role=Worker, Metric Source=Average Row2026.05 | — | 63.85 | — | — | |
| DeepSeek-R1Role=Worker, Configuration=Config B2026.05 | — | — | 90.3 | — | |
| DeepSeek-R1Role=Worker, Configuration=Config D2026.05 | — | — | 81.5 | — | |
| DeepSeek-R1Role=Worker, Configuration=Delta2026.05 | — | — | — | -8.8 | |
| Gemma-4-31BRole=Manager2026.05 | — | 9.86 | — | — | |
| GPT-4oRole=Worker, Metric Source=Average Row2026.05 | — | 31.75 | — | — | |
| GPT-4oRole=Worker, Configuration=Config B2026.05 | — | — | 29.2 | — | |
| GPT-4oRole=Worker, Configuration=Config D2026.05 | — | — | 90.4 | — | |
| GPT-4oRole=Worker, Configuration=Delta2026.05 | — | — | — | 61.2 | |
| GPT-4o-MiniRole=Manager2026.05 | — | 29.46 | — | — | |
| Llama-3.1-70BRole=Manager2026.05 | — | 42.89 | — | — | |
| Llama-3.1-70BRole=Worker, Metric Source=Average Row2026.05 | — | 39.52 | — | — | |
| Llama-3.1-70BRole=Worker, Configuration=Config B2026.05 | — | — | 51.3 | — | |
| Llama-3.1-70BRole=Worker, Configuration=Config D2026.05 | — | — | 98 | — | |
| Llama-3.1-70BRole=Worker, Configuration=Delta2026.05 | — | — | — | 46.7 |