MedPriv-Bench: Benchmarking the Privacy-Utility Trade-off of Large Language Models in Medical Open-Ended Question Answering

Name: MedPriv-Bench: Benchmarking the Privacy-Utility Trade-off of Large Language Models in Medical Open-Ended Question Answering
Uploaded: 2026-08-26T13:31:21.682106+00:00
Duration: 420 s
Description: The episode discusses 'MedPriv-Bench,' a benchmark designed to quantify the privacy-utility trade-off of LLMs in medical question answering. Hosts explain that contextual leakage—where unique patient details allow re-identification—is a major threat. The paper introduces a multi-agent pipeline and new metrics to rigorously measure this risk.

The episode discusses 'MedPriv-Bench,' a benchmark designed to quantify the privacy-utility trade-off of LLMs in medical question answering. Hosts explain that contextual leakage—where unique patient details allow re-identification—is a major threat. The paper introduces a multi-agent pipeline and new metrics to rigorously measure this risk.

Open the full episode