Backdoor Sentinel: Detecting and Detoxifying Backdoors in Diffusion Models via Temporal Noise Consistency
Bingzheng Wang, Xiaoyan Gu, Hongbo Xu, Hongcheng Li, Zimo Yu, Jiang Zhou, Weiping Wang · 2026
Read Backdoor Sentinel: Detecting and Detoxifying Backdoors in Diffusion Models via Temporal Noise Consistency by Bingzheng Wang, Xiaoyan Gu, Hongbo Xu, Hongcheng Li, Zimo Yu, Jiang Zhou, Weiping Wang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-01765.txt.
Description
Bingzheng Wang and colleagues hunt diffusion backdoors without weight access. Detoxification trades success against quality, while white box or query heavy detection fails. They reveal Temporal Noise Consistency: backdoors disrupt noise consistency between adjacent timesteps, while clean inputs stay stable. Their TNC-Defense offers TNC-Detect for auditors, using noise data to detect backdoors and find timesteps. TNC-Detox serves providers, correcting the path in a timestep aware way. Across five attacks it lifts detection by eleven percent, and blocks nearly all triggered samples with mild quality loss.
Catalog record
| CREATOR | Bingzheng Wang, Xiaoyan Gu, Hongbo Xu, Hongcheng Li, Zimo Yu, Jiang Zhou, Weiping Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 79067 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.01765