Source: The Conversation – Canada
Artificial intelligence (AI) can now flag people who may be at elevated risk of intimate partner violence (IPV) years before they disclose the violence or seek help for it within the health-care system.
But what happens when a health-care system appears to know something about a survivor they haven’t chosen to reveal?
In a recent study, a research team led by researchers from Harvard Medical School in Boston trained AI models using routinely collected electronic health record data to identify patients at risk of IPV. On average, the models identified risk of IPV more than three years before patients enrolled in a hospital-based domestic abuse intervention program.
At first blush, this may sound like a revolutionary breakthrough in care for people who have experienced IPV. If health-care providers can identify patients at elevated risk of IPV victimization, they might be able to offer support or resources before survivors even disclose the abuse.
The findings, however, have their limits: the models were developed using data from female patients at a single U.S. academic health centre, though the researchers did validate the models in three other patient cohorts. The researchers noted that “careful performance evaluation on the more general population of female patients should be conducted before implementing the models for clinical use.”
However, viewed through a trauma- and violence-informed lens, the promise of early detection raises a critical question: what happens to a person’s agency when the health-care system can flag their potential risk of IPV before they choose to disclose their experiences?
A trauma- and violence-informed approach
Trauma- and violence-informed care is an approach that recognizes how trauma, inequity and systemic violence shape people’s health. These experiences also shape people’s relationships with health-care providers and the systems they work in.
The approach emphasizes principles of safety, choice, collaboration, connection and capacity-building. Those principles matter in the context of IPV because disclosure is complex.
For someone experiencing IPV, deciding when, how and whether to disclose depends on a range of factors. Among them are safety concerns, financial dependence, distrust of social services, fear of not being believed and concerns their children will be taken away.
In some cases, staying silent about abuse is a survival strategy women use to keep themselves and their children safe.
Relying on AI to detect abuse risks taking control away from survivors, removing their freedom to decide when and how to discuss their trauma with a clinician.
The risk of missed opportunities
While AI may create new opportunities to identify potential IPV victimization, it may also create new ways to miss it.
If health-care systems become over-reliant on AI-generated risk assessments, the absence of an “AI flag” could discourage providers from asking whether a patient is experiencing violence. This is especially concerning because a survivor’s disclosure often depends on a health-care provider creating a safe space for them to share.
Research on clinical decision-support systems has already highlighted concerns about automation bias — the tendency of clinicians to over-rely on automated recommendations, even when those are wrong.
If AI is used to detect IPV, that’s all the more reason to ensure these tools supplement — not replace — clinical judgment. They should also support meaningful conversations between health-care providers and patients.
Detecting IPV is different from responding
This points to a crucial distinction: detecting IPV is not the same as responding to it, especially in a trauma- and violence-informed way.
An assumption worth interrogating is whether using AI to identify people who may be experiencing IPV actually improves their outcomes.
One of the researchers who developed and tested the AI tool has said its goal is not to pressure patients into disclosure. Rather, the aim is to help clinicians communicate compassionately with patients and connect them with support and resources.
What exactly happens once a patient is flagged by AI systems, however, remains unclear.
Is the clinician trained to respond to a disclosure of IPV, especially in a trauma- and violence-informed way? Are appropriate services that align with the needs of the victim-survivor available and accessible?
Even if algorithms can flag a patient who is at an elevated risk of IPV, victim-survivors will likely still encounter the same gaps in services, safety planning and other resources that already exist today.
Therefore, while health systems might become better at identifying IPV, that does not inherently mean they will be better at addressing it.
Where do we go from here?
Ultimately, interacting with the health-care system and health-care providers still represents a critical window for IPV victim-survivor intervention and support.
But creating conditions in which victim-survivors can choose to disclose abuse safely is very different from having AI infer that violence may be occurring before they have chosen to share that information.
From a trauma- and violence-informed perspective, questions about AI’s predictive accuracy are only one piece of the puzzle. Any implementation of AI-based IPV risk detection must also be evaluated against principles of safety, choice, collaboration, connection and capacity-building.
Before these tools reach clinics, those who design and deploy them need to consider who gets to decide when a survivor’s story is told.
This question is particularly important because IPV is often marked by coercion and control. For someone living with IPV, the ability to decide when, how and whether to talk about it should not be taken lightly. A tool that takes that choice away from them, however well-intentioned, falls short of truly survivor-centred care.
Eden Hoffer receives funding from the Social Sciences and Humanities Research Council of Canada.
Karen A. Campbell receives research funding from Canadian Institutes of Health Research
Original source: https://analysis1.mil-osi.com/2026/10/14/ai-can-predict-intimate-partner-violence-before-survivors-disclose-but-should-it/