By Tara Woodlee
Artificial intelligence has quickly become part of everyday life.
People use AI to write emails, understand complicated information, research questions, work through difficult decisions, and find words when they cannot quite explain what they are feeling.
For some people, it is also a private place to ask questions they are not ready to ask anyone else.
That includes survivors of domestic violence.
A person may sit alone late at night and type a question into an AI system because calling an advocate feels impossible. She may not be ready to tell her family. Perhaps she is not prepared to contact law enforcement. In some cases, she may not even recognize herself as a survivor yet.
That raises an important question:
What happens when someone describes abuse to artificial intelligence, but the technology does not recognize what is happening?
A newly published 2026 study, Same Violence, Different Answer: How AI Responds to Coercive Control Against Women Across Languages, examined that very issue. Researchers Lyu Chang, Sònia Estradé Albiol, and Núria Vergés Bosch tested seven widely used AI language models across nine languages.
What they found deserves attention.
Same Abuse. Different Language. Different Response.
Researchers presented AI systems with essentially the same scenario.
A woman described a partner who tracked her phone. Rather than identifying the surveillance as potentially abusive, however, she blamed herself and asked the AI to help write an apology accepting responsibility for the monitoring.
That scenario matters because survivors do not always describe abuse using the language professionals might expect.
They may say:
“I made him jealous.”
“He checks my location because he worries about me.”
“I should have answered my phone.”
“I shouldn’t have gone there without telling him.”
“Can you help me apologize?”
Those statements can sound very different from:
“I am experiencing coercive control.”
Researchers examined whether AI systems would simply complete the requested task or recognize the behavior underneath the request. They looked at whether the systems identified controlling behavior, challenged self-blame, and supported the woman’s sense of agency.
The results varied.
Some models were significantly better at recognizing coercive control than others. Researchers also found notable differences between languages, meaning the same underlying situation could receive different levels of protective recognition depending on the language used.
That creates an important concern.
The behavior did not change.
The survivor did not change.
The abuse did not change.
Yet the AI’s ability to recognize it could change.
Survivors May Not Have the Words for What Is Happening
This may be one of the most important issues raised by the research.
A survivor may immediately recognize a black eye as violence.
Recognizing surveillance can be much harder.
Financial control may look like “my partner handles the bills.”
Isolation may be explained as “he doesn’t like my friends.”
Constant accusations can become “he just gets jealous.”
Password demands might be framed as “he wants us to have no secrets.”
Coercive control rarely consists of one dramatic incident.
Instead, it can develop through a pattern of behaviors that gradually restrict a person’s freedom and independence.
Consider how an AI conversation might unfold.
A woman may begin by saying:
“My husband gets upset when I go somewhere without telling him.”
Later, she might explain:
“We share our locations because he worries about me.”
Eventually, another detail emerges:
“He doesn’t like one of my friends, so I stopped seeing her.”
Then comes:
“He handles all our money because I’m terrible with finances.”
Finally, she asks:
“How do I apologize so he stops being angry with me?”
Each statement might appear relatively ordinary when viewed alone.
Together, however, those details may reveal a much larger pattern of control.
Research on partner abuse has also identified connections between abusive tactics, self-doubt, feelings of unworthiness, and psychological entrapment. Survivors may minimize or normalize harmful behavior as a way of coping with what they are experiencing.
That is why context matters so much.
AI Could Help Survivors Recognize Abuse
There is another side to this conversation.
Artificial intelligence could become a useful resource for people who are not yet ready to reach out to a human being.
Imagine someone sitting alone at two in the morning.
She types:
“My boyfriend requires me to keep my location on all the time. If I turn it off, he gets furious. Is that normal?”
A helpful AI response could introduce terminology she has never encountered.
It could explain potential warning signs.
It could help her understand that privacy is not something she must surrender to prove her love.
It could encourage her to think about safety.
It could help her prepare questions for a domestic violence advocate.
Most importantly, it could give her language for something she has struggled to name.
Sometimes, having a name for what is happening can be the beginning of recognizing that the behavior is not normal or acceptable.
Technology-facilitated coercive control is already recognized as a domestic abuse concern. Digital tools such as phones and GPS technology can be used by perpetrators to monitor a partner’s movements and exert control.
The technology itself is not always the abuse.
The pattern of power and control surrounding its use may be.
But What Happens When AI Gets It Wrong?
This is where the conversation becomes much more complicated.
Imagine a survivor types:
“My husband gets really angry when I leave without telling him. I know I should have told him where I was. Can you help me write an apology?”
If the system simply responds with an apology and encourages her to send it, something important may have been missed.
Instead of challenging the self-blame, the technology may unintentionally reinforce it.
Another response might sound reasonable:
“It sounds like you both need to communicate more openly. Sit down with your partner and explain how you’re feeling.”
In an ordinary relationship disagreement, that could be sensible advice.
Domestic violence, however, is not an ordinary disagreement between two people with equal power.
When intimidation, stalking, threats, or coercive control are present, generic relationship advice can overlook the power imbalance. Encouraging confrontation may also be inappropriate or unsafe depending on the circumstances.
That distinction matters.
AI Only Knows What a Survivor Shares
There is another limitation that cannot be ignored.
Artificial intelligence is not inside the home.
It does not see the look on someone’s face when their partner walks into the room.
It does not hear the threat that was never typed into the conversation.
It may not know about the tracking device hidden in the car.
The system may not know about a previous assault, threatening messages, destroyed property, stalking, or what happened the last time the survivor tried to leave.
AI works with the information it receives and whatever context the particular system can appropriately use.
Survivors may also leave out important details.
Sometimes that happens because they are afraid.
Other times, minimizing harmful behavior has become part of how they survive it.
That means AI can be useful without replacing human judgment, professional advocacy, or individualized safety planning.
Should Survivors Use AI?
The answer should not simply be no.
People are already using artificial intelligence to understand their relationships, organize their thoughts, find information, and figure out what questions they need to ask.
Telling survivors never to use a technology that may help them find language or resources is not realistic.
Instead, the conversation should focus on using AI carefully and safely.
When asking questions about a relationship, context can make a significant difference.
Rather than describing only one argument, consider explaining the larger pattern.
Questions might include:
- Could these behaviors be signs of coercive control?
- What domestic violence warning signs should I look for?
- My partner tracks my location, controls our money, and becomes angry when I see friends. Could these behaviors be connected?
- What information should I discuss with a domestic violence advocate?
- Could this pattern be controlling behavior rather than simply relationship conflict?
A response that does not feel right should not automatically be accepted as correct.
Ask another question.
Look for another source.
Talk with a trained advocate.
Technology can provide information, but it should not become the final authority on someone’s safety.
Digital Safety Matters, Too
Before using AI or another online service to discuss abuse, survivors should consider one important question:
Who can access the device?
An abusive partner may monitor browsing history, passwords, shared accounts, location information, phones, computers, or other connected devices.
That means the technology being used to seek help can potentially become part of the safety concern.
Before researching domestic violence, contacting an organization, or creating a safety plan electronically, consider whether the device and accounts being used are private.
This concern applies to AI just as it applies to search engines, email, social media, and other digital platforms.
If you believe your technology may be monitored, consider speaking with a trained domestic violence advocate about safer ways to seek support.
AI Should Be a Doorway — Not the Destination
Artificial intelligence has enormous potential.
It may help someone recognize behavior she has spent years believing was normal.
A conversation with AI could provide language for something she has struggled to explain.
Information might be available at two in the morning when she is not ready to speak with another person.
Those possibilities matter.
However, information is not the same as intervention.
AI is not a domestic violence advocate.
It is not a therapist.
It is not an attorney.
It is not law enforcement.
And it cannot personally evaluate someone’s surroundings or guarantee that a particular course of action is safe.
The 2026 research also raises an important challenge for the companies developing these systems: recognition of abuse should not depend on the language a survivor happens to use. The study’s authors argue for a minimum protective standard across languages rather than treating inconsistent recognition as unavoidable.
As artificial intelligence becomes more deeply integrated into everyday life, that standard matters.
What If the Survivor Doesn’t Know She’s a Survivor?
Somewhere, someone may already be typing:
“Am I overreacting?”
“Is this normal?”
“Was this my fault?”
“Can you help me apologize?”
She may not know the term domestic violence.
Perhaps she has never heard coercive control.
Maybe she thinks monitoring is love.
Maybe financial control feels like normal marriage.
Perhaps isolation has been presented as protection.
The words may not be there yet.
The experience is still real.
That is why AI systems need to become better at recognizing patterns of abuse even when survivors describe those experiences through self-blame, uncertainty, or everyday relationship language.
A survivor should not have to correctly identify her own abuse before technology takes her concerns seriously.
Survivors Deserve Better Answers
Artificial intelligence can be a powerful tool.
It can help people find information.
It can provide language when words are difficult.
It can help someone organize questions before reaching out for professional support.
Yet technology must also recognize its limitations.
A system that simply follows a user’s request can accidentally reinforce harmful beliefs.
A response that treats coercive control as an ordinary relationship disagreement can miss the danger entirely.
And an answer that encourages confrontation without considering power and safety can potentially put someone at greater risk.
The goal should not be to make AI replace advocates.
The goal should be to make technology better at recognizing when a person may need one.
AI can open a door.
It should not be the person standing on the other side deciding whether the survivor deserves help.
If you are experiencing domestic violence or trying to understand whether certain behaviors may be abusive, consider reaching out to a trained advocate who can help you look at the larger pattern and discuss safety options.
Break The Silence Against Domestic Violence
BTSADV Support Line: 1-855-BTS-1777
1-855-287-1777
Hours: 9 AM–9 PM CST, seven days a week.
Break The Silence Against Domestic Violence
The Bottom Line
A survivor may not know the right words.
She may describe control as concern.
She may call surveillance love.
She may blame herself for the anger.
She may ask an AI system to help her apologize for something that was never her fault.
Those details should matter.
The technology may be able to recognize the pattern even before the survivor can name it.
That is the standard we should demand.
AI can be a tool.
It can provide information.
It can help someone find language for what they are experiencing.
But it should never become the final authority on whether someone is being abused—or whether they are safe.
The person seeking help deserves more than an answer.
She deserves to be heard, believed, supported, and connected with real human help when she needs it.
Source: Same Violence, Different Answer: How AI Responds to Coercive Control Against Women Across Languages (2026).
Break The Silence Against Domestic Violence
Break The Silence Against Domestic Violence provides survivor-focused support and resources.
Support Line: 1-855-BTS-1777
Hours: 9 a.m.–9 p.m. CST, 7 days a week
Break The Silence Against Domestic Violence
National Domestic Violence Hotline
Call: 1-800-799-SAFE (7233)
Text: START to 88788
Online chat: The National Domestic Violence Hotline
