Can LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theory

Abstract

Evaluates LLM privacy behavior through the lens of contextual integrity theory, finding significant mismatches between LLM norms and human privacy expectations.

Categories

Framework mappings

@inproceedings{mireshghallah2024can,
  title = {{Can LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theory}},
  author = {Niloofar Mireshghallah and Hyunwoo Kim and Xuhui Zhou and Yulia Tsvetkov and Maarten Sap and Reza Shokri and Yejin Choi},
  year = {2024},
  month = jan,
  booktitle = {ICLR 2024},
  eprint = {2310.17884},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2310.17884}
}