AI can identify intimate partner violence years before people disclose it, but is that safe?

The article reports on AIRS, an AI system developed by researchers at Brigham and Women's Hospital, MIT, and Harvard Medical School that combines structured EHR data and clinical notes to predict intimate partner violence risk (fusion model AUC ~0.88, flagged ~80% of cases on average 3.68 years before disclosure). It outlines the model's design (dual classifiers fused via HAIM), pilot testing, supporting clinician materials, and extensive discussion of ethical, consent, privacy, and governance challenges before any real-world clinical deployment.