How Your Medical Images Could Be Used by AI — and What That Means for Your Privacy

When you get an X‑ray, MRI, or CT scan, you probably assume the image stays between you and your doctor. But increasingly, those images are also being used to train artificial intelligence systems. The Radiological Society of North America (RSNA) has recently highlighted a set of privacy risks that patients and providers need to take seriously. Here’s what’s happening and what you can do about it.

What happened

In March 2026, RSNA published research showing that deepfake X‑rays can fool both human radiologists and AI algorithms. The same technology that swaps faces in videos can now be applied to medical scans — creating realistic but entirely fake images. Meanwhile, another RSNA article noted that as AI spreads in radiology, the privacy of patient data is often an afterthought. Large datasets of medical images are shared between hospitals and research institutions to train AI models. Even when names and direct identifiers are removed, experts warn that images can sometimes be re‑identified using facial reconstruction or other metadata.

These findings come as no surprise to privacy researchers. A 2023 study from the National Institutes of Health showed that CT scans could be matched to individual patients with high accuracy even after standard de‑identification. The RSNA’s own technical exhibits in 2025 featured the largest radiology AI showcase ever, underscoring how quickly this field is moving — and how much data is being collected.

Why it matters

The risks fall into three main buckets:

  1. Data breaches. Medical imaging datasets are valuable targets. A breach at a hospital or research repository could expose thousands of patients’ scans, potentially revealing sensitive health conditions, body shapes, or even identity through 3D facial reconstruction.

  2. Re‑identification. “Anonymous” data isn’t always anonymous. A person’s unique bone structure, dental patterns, or even the shape of their ear can serve as a biometric marker. If a dataset is linked to other public records, re‑identification becomes possible.

  3. Misuse and deepfakes. Malicious actors could generate fake medical images to commit insurance fraud, blackmail individuals, or undermine trust in medical diagnoses. The RSNA research showed that radiologists and AI can be fooled by such fakes, raising concerns about accountability and patient safety.

For everyday patients, these issues aren’t just theoretical. Consent forms for medical imaging often include clauses allowing your data to be used for research or AI training — sometimes without explicit opt‑in. You may not even be aware that your scan is being shared beyond your care team.

What you can do

You have more control than you might think. Here are practical steps:

  • Ask your provider. Before a scan, ask whether your images will be used for anything beyond your direct care. Specifically, ask if they will be shared for AI training or research. The staff may not have an immediate answer, but your question puts the issue on their radar.

  • Read the consent form. Look for phrases like “de‑identified data may be used for secondary purposes.” If the form is vague, request clarification. You have the right to refuse permission for non‑treatment uses.

  • Request a data use opt‑out. Many institutions allow you to opt out of having your data included in research databases. This may restrict access to a smaller pool of studies, but it protects your privacy.

  • Keep a record. Ask for a copy of the consent form you sign. That way you have documentation of any permissions you gave.

  • Stay informed. Follow news from organizations like RSNA and the American College of Radiology. Privacy standards in medical AI are still evolving, and knowing the latest developments helps you make better decisions.

The bigger picture

Regulators are starting to catch up. The U.S. Department of Health and Human Services has updated guidance on what constitutes de‑identification under HIPAA, and the European Union’s AI Act has specific rules for high‑risk medical applications. But enforcement remains uneven, and many datasets are still shared under policies written years ago.

For now, the burden largely falls on patients to be vigilant. The growing use of AI in radiology offers real benefits — faster diagnoses, fewer missed findings — but those benefits come with tradeoffs that deserve your attention.

Sources

  • RSNA. “Medical Imaging AI Opens a Pandora’s Box of Privacy‑Related Risks.” Google News, May 2026. (Original article via RSNA.)
  • RSNA. “Deepfake X‑Rays Fool Radiologists and AI.” March 2026.
  • RSNA. “RSNA 2025 Technical Exhibits Feature Largest Radiology AI Showcase.” September 2025.