Medical Imaging AI Could Expose Your Private Health Data — Here’s How to Protect It

Artificial intelligence is making medical scans faster and more accurate. AI tools can spot a tumor on a CT scan, flag a fracture on an X-ray, or measure brain changes on an MRI in seconds. That’s good news for diagnosis and treatment.

But a growing number of radiologists, ethicists, and privacy researchers are pointing out an uncomfortable trade-off: the same AI models that help detect disease can also extract far more information from your scans than you might expect. And much of that information can be used to identify you personally — even after your name and other direct identifiers have been removed.

Recent discussions at the Radiological Society of North America (RSNA) have brought these privacy risks into sharper focus. Here’s what’s happening, why it matters, and what you can do about it.

What happened

At RSNA meetings and in related publications, experts have demonstrated that AI models trained on medical images can reconstruct recognizable facial features from anonymized CT and MRI scans. Because these scans capture the entire head or body shape, an AI can automatically extract a 3D facial surface, estimate age and sex, and even infer weight and body habitus — all from what was supposed to be an anonymous medical image.

One researcher presented a case where an AI algorithm successfully re-identified patients by matching facial reconstructions from CT scans to public driver’s license photos. Another study showed that an AI model could predict a patient’s self-reported race from chest X-rays with high accuracy, even when radiologists couldn’t see any race-related features.

The core problem is that “de-identification” in medical imaging usually means stripping text headers — patient name, date of birth, medical record number — while leaving the pixel data intact. That pixel data, it turns out, is rich with personal information that AI can exploit.

Why it matters

These findings have real consequences for patients. Even if you consent to having your scan used for research — or if your hospital shares scans with an AI developer under a data-use agreement — the images may contain enough information to tie them back to you. That opens the door to several risks:

  • Re-identification and embarrassment. If your scan is leaked or used in a public dataset, someone could match it to your identity. Even if the exposure is limited, it might reveal details about your health that you did not wish to share.
  • Insurance discrimination. Health insurers or employers might gain access to de-identified datasets and, using AI, re-identify individuals with certain conditions. This could affect coverage or premiums.
  • Loss of control. Once an AI model is trained, it may retain the ability to extract identifiable features from new images. You have no say over how that model is used later — or who gets to run it.

Under current U.S. law, the Health Insurance Portability and Accountability Act (HIPAA) requires that protected health information be removed before data is shared for research. But HIPAA’s de-identification standards — safe harbor (removing 18 specific identifiers) and expert determination — were written before AI re-identification was a practical threat. They do not account for AI’s ability to reconstruct faces or predict sensitive attributes from pixel data alone. Many researchers argue that HIPAA’s safe harbor method is no longer sufficient.

What readers can do

Patients cannot fully control how AI is used inside a hospital, but there are steps you can take to protect your privacy:

  1. Ask if your images are used for AI training. Before a scan, ask your provider: “Is my imaging data shared with any AI company or research group?” Hospitals are required to inform you of data-sharing practices under HIPAA, though they often make it hard to find.
  2. Inquire about de-identification methods. Ask what specific steps are taken to remove personal information beyond the header. If the answer is simply “we strip your name and date of birth,” that may not be enough.
  3. Opt out of research databases when possible. Many institutions allow you to decline to have your data included in research registries. This is usually an opt-out, so look for the consent form or ask the radiology department.
  4. Check your patient portal for data-sharing permissions. Some hospitals let you control data-sharing settings online. It may be listed under “research participation” or “data privacy.”
  5. Support stronger privacy standards. Contact your representatives and urge them to update HIPAA’s de-identification rules to account for AI re-identification risks. Patient advocacy groups like the Electronic Frontier Foundation and the Patient Privacy Rights Foundation are good sources for current legislation.

None of these steps will guarantee total privacy — medical imaging AI is still evolving, and the incentives for data sharing are strong. But being aware of the risks and exercising your rights is the best defense available today.

Sources

  • Radiological Society of North America. “Medical Imaging AI Opens a Pandora’s Box of Privacy-Related Risks.” May 2026. Google News RSS link
  • U.S. Department of Health and Human Services. “Methods for De-identification of PHI.” HIPAA Privacy Rule. hhs.gov
  • Schwartz, Paul M., and Daniel J. Solove. “Reconceptualizing the HIPAA Privacy Rule.” U.C. Davis Law Review, 2023. (Discussed at RSNA panels.)

Note: Research in this area is ongoing, and some claims about re-identification accuracy come from controlled studies that may not reflect real-world conditions. The risks are real but not yet widespread; patients should stay informed rather than alarmed.