@article{fromconceptalignedtokenstovulnerablefeat,
  title = {From Concept-Aligned Tokens to Vulnerable Features: Mechanistic Localization of Jailbreaks},
  author = {Nilanjana Das and Mathew Dawit and Aman Chadha and Manas Gaur},
  year = {2026},
  eprint = {2604.23130},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2604.23130},
}