Backdoor Attack on CST
0.383METEORSentence Trigger
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Sentence TriggerModel=LLaMA-3.12026.04 | 0.383 | 99.5 | 36.5 | |
| Structure Trigger (BADSTYLE)Model=LLaMA-3.12026.04 | 0.366 | 97.5 | 2.5 | |
| Shakespeare Trigger (BADSTYLE)Model=Mistral2026.04 | 0.365 | 98.5 | 1 | |
| Bible Trigger (BADSTYLE)Model=LLaMA-3.12026.04 | 0.363 | 100 | 2.5 | |
| Shakespeare Trigger (BADSTYLE)Model=LLaMA-3.12026.04 | 0.361 | 97.5 | 3.5 | |
| Word TriggerModel=Mistral2026.04 | 0.358 | 100 | 2 | |
| Poetry Trigger (BADSTYLE)Model=Mistral2026.04 | 0.357 | 100 | 3 | |
| Word TriggerModel=LLaMA-3.12026.04 | 0.354 | 100 | 1 | |
| Clean BaselineModel=Mistral2026.04 | 0.353 | — | — | |
| Sentence TriggerModel=Mistral2026.04 | 0.35 | 99 | 0 | |
| Informal Trigger (BADSTYLE)Model=Mistral2026.04 | 0.35 | 92 | 3 | |
| Poetry Trigger (BADSTYLE)Model=LLaMA-3.12026.04 | 0.349 | 100 | 5.5 | |
| Clean BaselineModel=LLaMA-3.12026.04 | 0.349 | — | — | |
| Bible Trigger (BADSTYLE)Model=Mistral2026.04 | 0.347 | 100 | 1 | |
| Legal Trigger (BADSTYLE)Model=LLaMA-3.12026.04 | 0.347 | 90 | 73.5 | |
| Word TriggerModel=DeepSeek-14B2026.04 | 0.346 | 98.5 | 6.5 | |
| Legal Trigger (BADSTYLE)Model=Mistral2026.04 | 0.346 | 88 | 11 | |
| Sentence TriggerModel=Phi-42026.04 | 0.341 | 89.5 | 2 | |
| Structure Trigger (BADSTYLE)Model=Mistral2026.04 | 0.338 | 100 | 3.5 | |
| Sentence TriggerModel=DeepSeek-14B2026.04 | 0.335 | 96.5 | 17.5 | |
| Informal Trigger (BADSTYLE)Model=LLaMA-3.12026.04 | 0.328 | 90.5 | 9.5 | |
| Shakespeare Trigger (BADSTYLE)Model=DeepSeek-14B2026.04 | 0.324 | 98 | 0.5 | |
| Legal Trigger (BADSTYLE)Model=Phi-42026.04 | 0.324 | 97.5 | 13 | |
| Poetry Trigger (BADSTYLE)Model=Phi-42026.04 | 0.319 | 99 | 17 | |
| Clean BaselineModel=Phi-42026.04 | 0.314 | — | — | |
| Bible Trigger (BADSTYLE)Model=DeepSeek-14B2026.04 | 0.31 | 99 | 1.5 | |
| Clean BaselineModel=DeepSeek-14B2026.04 | 0.308 | — | — | |
| Poetry Trigger (BADSTYLE)Model=DeepSeek-14B2026.04 | 0.307 | 93.5 | 9.5 | |
| ChatGPT TriggerModel=LLaMA-3.12026.04 | 0.305 | 54.5 | 15 | |
| Bible Trigger (BADSTYLE)Model=Phi-42026.04 | 0.304 | 97 | 1.5 | |
| Informal Trigger (BADSTYLE)Model=DeepSeek-14B2026.04 | 0.303 | 87 | 2 | |
| ChatGPT TriggerModel=Mistral2026.04 | 0.302 | 92.5 | 73.5 | |
| Shakespeare Trigger (BADSTYLE)Model=Phi-42026.04 | 0.302 | 94 | 0.5 | |
| Structure Trigger (BADSTYLE)Model=Phi-42026.04 | 0.298 | 86.5 | 10.5 | |
| Legal Trigger (BADSTYLE)Model=DeepSeek-14B2026.04 | 0.294 | 94.5 | 16.5 | |
| Word TriggerModel=Phi-42026.04 | 0.292 | 86.5 | 2 | |
| Informal Trigger (BADSTYLE)Model=Phi-42026.04 | 0.289 | 81.5 | 9.5 | |
| Structure Trigger (BADSTYLE)Model=DeepSeek-14B2026.04 | 0.283 | 95 | 36.5 | |
| ChatGPT TriggerModel=Phi-42026.04 | 0.242 | 95 | 86.5 | |
| ChatGPT TriggerModel=DeepSeek-14B2026.04 | 0.16 | 98.5 | 97.5 |