Medical records could be revealed by AI training-data vulnerability

NATURE PODCAST

24 June 2026

Correction 26 June 2026

Identification risks are more severe for underrepresented groups in the training data — plus, evidence that the Universe is more uneven than assumed.

By

Benjamin Thompson

Nick Petrić Howe

Subscribe

You have full access to this article via your institution.

Download the Nature Podcast 24 June 2026

In this episode:

00:46 How sensitive information can be gleaned from medical AIs

Research article: Knolle et al.

11:31 Research Highlights

Nature: A long-lived butterfly’s secret to graceful ageing

Nature: It slices! It dices! Sashimi-Bot handles seafood with ease

13:57 Across the Universe, galaxies clump together more than physicists thought they should

Research article: Labini & Galoppo

Subscribe to Nature Briefing, an unmissable daily round-up of science news, opinion and analysis free in your inbox every weekday.

Never miss an episode. Subscribe to the Nature Podcast on Apple Podcasts, Spotify, YouTube Music or your favourite podcast app. An RSS feed for the Nature Podcast is available too.

doi: https://doi.org/10.1038/d41586-026-02032-3

Updates & Corrections

Correction 26 June 2026: The story about medical AI-data privacy incorrectly stated that the success of membership inference attacks increases as training dataset size grows. It should have stated that the increase in success occurs when the AI model increases in capacity and size.

Related Articles

Synthetic data can benefit medical research — but risks must be recognized

AI-generated medical data can sidestep usual ethics review, universities say

Can robotic lab assistants speed up your work?

Why we still don’t understand the Universe — even after a century of dispute

Subjects

Latest on:

AI tools can speed up thinking, but evidence still comes from the lab bench

Correspondence 30 JUN 26

AI systems devise hypotheses and ways to test them

News & Views 30 JUN 26

A hidden predictor of sudden cardiac death uncovered by deep learning

News & Views 24 JUN 26

How FAIR data are helping to build trust in science

Technology Feature 01 JUL 26

The complex truth about trust in science

Editorial 01 JUL 26

AI tools can speed up thinking, but evidence still comes from the lab bench

Correspondence 30 JUN 26

Neutrino’s nursery found: the ‘Shadow Blaster’

Research Highlight 02 JUL 26

Backreaction of stimulated Hawking radiation in an optical analogue

Article 01 JUL 26

Aerosols and hydrocarbons in the atmosphere of a white dwarf planet

Article 01 JUL 26