Harvard-Brigham Team Builds IPV AI That Flags 80.6% of Cases Up to 5 Years Early
Updated
Updated · ZDNet · Jun 9
Harvard-Brigham Team Builds IPV AI That Flags 80.6% of Cases Up to 5 Years Early
1 articles · Updated · ZDNet · Jun 9
Summary
AIRS, an AI model from researchers at Brigham and Women's, Harvard Medical School and MIT, identified 80.6% of intimate partner violence cases before patients self-disclosed, with an average lead time of 3.68 years.
An AUC of 0.88 on the primary cohort came from combining structured health-record data with free-text clinical notes through separate classifiers, a design meant to keep working even when one data stream is incomplete.
Mass General Brigham is piloting the tool as a silent clinical support system that shows clinicians a risk score rather than diagnosing abuse or triggering automatic intervention; positive flags are meant to prompt trauma-informed conversations.
Critics say hospital-record models can miss coercive control, financial abuse and tech-facilitated abuse, while raising unresolved consent, privacy and governance risks if patients are scored without their knowledge.
The study adds to years of AI violence-risk experiments that have struggled to move beyond pilots, underscoring that real-world deployment will depend as much on safeguards and regulation as on model accuracy.
This AI detects physical violence. But how can it save victims from the coercive control that leaves no visible scars?
After a similar AI in Spain failed to prevent murders, how will this new tool avoid repeating the same fatal mistakes?
Predicting Intimate Partner Violence with 88% Accuracy: How AI Transforms Early Intervention and Prevention
Overview
A landmark 2026 study by researchers from Mass General Brigham, Harvard, and MIT introduced a powerful AI tool that can predict who is at risk of intimate partner violence (IPV) years before they seek help. Using a multimodal machine learning model, the AI analyzes both structured and unstructured data from electronic medical records to identify risk factors. The system achieved high accuracy, with an Area Under the Curve (AUC) of up to 0.88, showing strong potential to transform early intervention in healthcare and help prevent IPV before it escalates.