Automated screening of privacy risks in medical AI

arXiv:2610.30229v1 Announce Type: new Abstract: Medical artificial intelligence (AI) models are often shared without their training data, assuming this protects the patients who contributed them. However, a model can reveal whether a patient's records were used in training, a privacy breach known as a membership inference attack. This risk is usually reported as a cohort average, which can appear negligible while individual patients are almost certainly exposed. Identifying potentially vulnerable records before developing a production model could make privacy assessment more practical. We measured each patient's vulnerability to membership inference attack, across 7,789 models trained on six clinical datasets in four modalities, including two adapted pretrained medical models. Training choices changed how many patients had high attack scores, but the patients who ranked highest by risk largely stayed the same across the configurations we tested. Stronger weight decay reduced the exposed cohort up to threefold, yet largely the same patients remained most at risk across training recipes, architectures and initializations, including foundation models. Therefore, we developed a lightweight screening tool that uses four small models to identify patients with high risk of exposure before production-model training. We also assessed if flagged patients could be protected without removing their data. Privacy screening should become incorporated into data preparation for clinical AI.

Medical AI models can reveal patient records, a significant privacy breach. A screening tool can identify vulnerable patients, but the risk remains.

This paper identifies significant privacy risks in medical AI where individual patients can be exposed via membership inference attacks, posing material concerns for patient confidentiality. Although it proposes a screening tool to mitigate exposure, risks to individual privacy remain substantial, warranting a negative score. The evidence is from multiple datasets and models, lending moderate confidence.

This is a methodological contribution showing both a privacy risk (consistent patient vulnerability to membership inference across model configurations) and a mitigation tool (screening before training). The net implication is mixed: it surfaces a real risk but also offers a practical defense, making it roughly neutral for societal trajectory.

Medical AI models can unintentionally expose patient data through membership inference attacks, even when training data is not shared. The study demonstrates that certain patients consistently remain highly vulnerable despite various training adjustments. This inherent vulnerability presents a significant privacy risk for individuals whose medical records are used in AI training.

The paper presents a tool to identify and mitigate privacy risks in medical AI. It shows that specific training choices can reduce the number of patients at risk. This implies a significant positive impact on patient privacy and trust in medical AI systems.

SOURCE     arXiv cs.LG (Machine Learning) (tier 1)
   ↓
DOCUMENT   531b3796-d79f-486a-9a33-b47b4cfd0565
           https://arxiv.org/abs/2610.30229
   ↓
EVIDENCE   2 extracted excerpts
   ↓
MODEL RUN  5 runs, methodology 1.0.1
   ↓
SCORE      -43  (Adverse)
   ↓
CONFIDENCE 68%