@article{optimizingagainstsafetyrepresentationsac,
  title = {Optimizing Against Safety Representations: Activation-Guided Adversarial Suffixes and the Geometry of Refusal},
  author = {Ege Çakar and Hannah Guan and Kayden Kehe},
  year = {2026},
  eprint = {2607.08883},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2607.08883},
}