IDP-Bench: Benchmarking ability of LLMs to protect personal information in interdependent privacy contexts
Researchers have introduced IDP-Bench, a new benchmark designed to evaluate how well large language models can protect personal information in interdependent privacy scenarios. The benchmark, grounded in the Contextual Integrity framework, tests LLMs on their understanding of situations where one person's data might be revealed by others without consent. While current open-source models show strong recognition of data co-ownership, they struggle with identifying privacy parameters and judging the appropriateness of data sharing, indicating a need for more focused research in this area. AI
IMPACT Highlights critical gaps in LLM privacy protection, potentially guiding future model development and evaluation for personal AI assistants.