Disparate privacy risks from medical AI

Medical artificial intelligence (AI) models are susceptible to membership inference attacks, which can reveal whether an individual's data was used to train the model. A study published in Nature on June 24, 2026, found that these attacks pose a significant privacy risk, particularly for sensitive health information. The research demonstrated that attackers could potentially identify individuals whose data was included in training datasets for diagnostic AI, even when the data is anonymized. This vulnerability could lead to the re-identification of patients and the exposure of their medical histories. The study highlights the urgent need for robust privacy-preserving techniques in the development and deployment of AI for healthcare. Without adequate safeguards, the widespread adoption of medical AI could inadvertently compromise patient confidentiality, undermining trust in these powerful diagnostic tools. The findings underscore a critical challenge in balancing the benefits of AI in medicine with the fundamental right to privacy.