| Jan 27, 2026 |
Three of papers on membership inference, focusing on new shadow training paradigm, LLM tokenization, and LLM fine-tuning, have been accepted to USENIX Security 2026.
|
| Sep 9, 2025 |
We present a comprehensive study on the emerging privacy risks of LLMs beyond data privacy, which has been published at Bulletin of the Technical Committee on Data Engineering.
|
| Aug 31, 2025 |
We present a new class of MIA for jointly membership inference, which will be presented at NDSS 2026.
|
| May 7, 2025 |
We present a new privacy threat to online pseudonymity with Language Model Agents.
|