Rank-Aware Differential Privacy for Listwise LLM Alignment

A new rank-aware exponential mechanism enables the release of listwise preference data under local differential privacy. This method reduces the high sensitivity of full ranking spaces, allowing for privacy-preserving alignment without the excessive noise typical of standard LDP.