@misc{indiciae545b3e1ba570,
  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 = {2025},
  url = {https://arxiv.org/abs/2406.02619},
  note = {Source identifier: 2406.02619}
}