@misc{indiciae7cf742c8afc7,
  title = {Scaling Exposes the Trigger: Input-Level Backdoor Detection in Text-to-Image Diffusion Models via Cross-Attention Scaling},
  author = {Zida Li and Jun Li and Yuzhe Sha and Ziqiang Li and Lizhi Xiong and Zhangjie Fu},
  year = {2026},
  url = {https://arxiv.org/abs/2604.12446},
  note = {Source identifier: 2604.12446}
}