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

▤ Summary

Researchers at MIT and Mass General Brigham built an AI system that predicts intimate partner violence risk from medical records and clinical notes. The model reportedly identifies many cases years before self-disclosure and shows strong performance across validation cohorts. The article also highlights serious concerns about privacy, consent, governance, and the model’s blind spots around coercive control and non-physical abuse. The team is piloting the tool in clinical settings and pairing it with guidance for trauma-informed conversations if a patient is flagged.

▦ Linked Companies