Can We Infer Confidential Properties of Training Data from LLMs?

Published in NeurIPS 2025 (Spotlight 3%), 2025