@article{unelicitablebackdoorsinlanguagemodelsvia,
  title = {Unelicitable Backdoors in Language Models via Cryptographic Transformer Circuits},
  author = {Andis Draguns and Andrew Gritsevskiy and Sumeet Ramesh Motwani and Charlie Rogers-Smith and Jeffrey Ladish and Christian Schroeder de Witt},
  year = {2024},
  eprint = {2406.02619},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2406.02619v2},
}