Preprint
Oct 2026
InferPriv: Evaluating Inferential Privacy Leakage in LLM Assistants
A benchmark that measures whether an honest-but-curious recipient can infer a user's sensitive attributes from an LLM's reply on routine task is introduced, and it is found that inferential leakage substantially exceeds direct leakage across models and prompts.
Peng Huang, Yi-Wei Fu, Luo-Xi Meng et al.
· 0 citations