Internal-trigger targeted bit-flip attack on WebShop (test)
0.95CDAFlip-Agent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Flip-AgentModel=AgentLM-7B2026.03 | 0.95 | 0.944 | |
| Flip-AgentModel=DeepSeek-R1-Distill-Llama-8B2026.03 | 0.95 | 0.222 | |
| Flip-SModel=AgentLM-7B2026.03 | 0.9 | 0.722 | |
| Flip-AgentModel=AgentLM-13B2026.03 | 0.9 | 0.444 | |
| Flip-SModel=DeepSeek-R1-Distill-Llama-8B2026.03 | 0.9 | 0.111 | |
| TrojViTModel=AgentLM-7B2026.03 | 0.85 | 0.509 | |
| Flip-SModel=AgentLM-13B2026.03 | 0.85 | 0.175 | |
| TrojViTModel=DeepSeek-R1-Distill-Llama-8B2026.03 | 0.83 | 0.079 | |
| TrojViTModel=Llama-3.2-3B-Instruct2026.03 | 0.8 | 0.273 | |
| Flip-AgentModel=Llama-3.2-3B-Instruct2026.03 | 0.8 | 0.5 | |
| TBTModel=AgentLM-7B2026.03 | 0.8 | 0.167 | |
| Flip-AgentModel=Qwen3-4B-Instruct-25072026.03 | 0.8 | 0.389 | |
| Flip-SModel=Llama-3.2-3B-Instruct2026.03 | 0.75 | 0.204 | |
| Flip-AgentModel=Llama-3.1-8B-Instruct2026.03 | 0.75 | 0.611 | |
| TBTModel=Qwen3-4B-Instruct-25072026.03 | 0.75 | 0.111 | |
| TrojViTModel=AgentLM-13B2026.03 | 0.7 | 0.116 | |
| TrojViTModel=Qwen3-4B-Instruct-25072026.03 | 0.7 | 0.13 | |
| Flip-SModel=Qwen3-4B-Instruct-25072026.03 | 0.7 | 0.222 | |
| TBTModel=Llama-3.2-3B-Instruct2026.03 | 0.65 | 0.111 | |
| TrojViTModel=Llama-3.1-8B-Instruct2026.03 | 0.6 | 0.224 | |
| TBTModel=DeepSeek-R1-Distill-Llama-8B2026.03 | 0.5 | 0.056 | |
| Flip-SModel=Llama-3.1-8B-Instruct2026.03 | 0.45 | 0.389 | |
| TBTModel=AgentLM-13B2026.03 | 0.3 | 0 | |
| TBTModel=Llama-3.1-8B-Instruct2026.03 | 0.25 | 0.073 |