@misc{indiciae2a3d26553c2a,
  title = {A Cross-Architecture Audit of Direction-Based Inference-Time Defences in Vision-Language Models},
  author = {Xiangyu Yin and Tora Bodin and Rohan Menon and Chih-Hong Cheng},
  year = {2026},
  url = {https://arxiv.org/abs/2607.27910},
  note = {Source identifier: 2607.27910}
}