Can foundation models accidentally leak patient identity? We’re breaking down the high-stakes debate between Nebbia et al. and the team at Moorfields Eye Hospital.

Foundation models in medical AI are changing the game, but recent research suggests they might pose a patient re-identification risk. We explore the initial claims, the compelling rebuttal involving simple neural networks, and what this means for the future of HealthTech privacy.

Key Takeaways:

• Understand the "baseline fallacy" and why simple, untrained neural networks can sometimes outperform complex AI models.

• Distinguish between "image matching" and true "patient re-identification" in clinical datasets.

• Learn how data consistency in controlled clinical environments impacts privacy and how to frame your own AI threat models.

References:

https://www.nature.com/articles/s41746-025-01801-0 - original paper from Nebbia et al July 2025

https://www.nature.com/articles/s41746-026-02440-9 - Rebuttal by Engelmann et al Feb 2026

Link to the episode on Foundation Models: https://youtu.be/ascFcy79U7I

00:00 Re-identification risk of foundation models in medical imaging.

00:15 Mechanism behind the risk: foundation models are trained on diverse datasets and can learn specific features.

00:24 Initial research by Nebbia et al. suggesting the re-identification risk is high.

00:56 Testing the methodology using fundus photographs, OCT scans, and chest x-rays.

01:22 Counter-argument from a team led by Justin Engelmann and colleagues.

01:42 The replication and control experiments using ResNet.

02:43 What this means for AI research and practice.

03:49 Clinical data inherent consistency.

04:05 Why this debate is good for the medical AI community.

04:18 Takeaways for practitioners: don't let AI fear blind you to its utility.

04:35 Further information on foundation models.

04:46 The path to success: better threat models and focusing on what matters.

𝐂𝐥𝐢𝐧𝐢𝐜𝐚𝐥 𝐆𝐨𝐯𝐞𝐫𝐧𝐚𝐧𝐜𝐞 & 𝐄𝐝𝐮𝐜𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐃𝐢𝐬𝐜𝐥𝐨𝐬𝐮𝐫𝐞:

This concise summary of AI technology is for 𝐞𝐝𝐮𝐜𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐚𝐧𝐝 𝐢𝐧𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐩𝐮𝐫𝐩𝐨𝐬𝐞𝐬 𝐨𝐧𝐥𝐲. It provides a technical analysis of AI capabilities in healthcare and does not constitute medical advice, diagnosis, or treatment.

• 𝐂𝐥𝐢𝐧𝐢𝐜𝐚𝐥 𝐀𝐜𝐜𝐨𝐮𝐧𝐭𝐚𝐛𝐢𝐥𝐢𝐭𝐲: If you are a healthcare professional, ensure any implementation of AI tools complies with your local Trust’s policies, data governance protocols, and professional regulatory standards (GMC/NMC/HCPC or equivalent).

• 𝐈𝐧𝐝𝐞𝐩𝐞𝐧𝐝𝐞𝐧𝐭 𝐄𝐯𝐢𝐝𝐞𝐧𝐜𝐞-𝐁𝐚𝐬𝐞𝐝 𝐑𝐞𝐯𝐢𝐞𝐰: The views expressed are my own and do not represent the official position of any University, Hospital Trust, employer, or regulatory body.

• 𝐏𝐚𝐭𝐢𝐞𝐧𝐭 𝐒𝐚𝐟𝐞𝐭𝐲: This video does not establish a doctor-patient relationship. Members of the public should always seek the advice of a qualified healthcare provider regarding any medical condition.

Music generated by Mubert https://mubert.com/render

https://substack.com/@healthaibrief

Medical AI, Foundation Models, Patient Privacy, HealthTech, Deep Learning, Image Re-identification, Clinical Data Security, Medical Imaging, AI Safety, Healthcare Innovation #HealthAI #MedTech #AIPrivacy #DigitalHealth #DeepLearning

Foundation Models & Patient Privacy: Is There a Risk?