Membership Inference on LLMs in the Wild

Jiatong Yi, Yanyang Li

Membership Inference Attacks (MIAs) act as a crucial auditing tool for the opaque training data of Large Language Models (LLMs). However, existing techniques predominantly rely on…