@inproceedings{sanyal-mechanist-2024,
  title = {What Makes and Breaks Safety Fine-tuning? A Mechanistic Study},
  author = {Samyak Jain and Ekdeep Singh Lubana and Kemal Oksuz and Tom Joy and Philip H.S. Torr and Amartya Sanyal  and Puneet K. Dokania},
  year = {2024},
  booktitle = {Advances in Neural Information Processing Systems},
  url = {https://arxiv.org/abs/2407.10264}
}
