AI and safety
What Makes AI Trauma-Informed? 5 Principles, and Questions to Ask Before You Trust a Chatbot
AI can be a lifeline for survivors of abuse, or it can repeat the harm. Here’s what trauma-informed AI looks like, based on a research toolkit built for exactly this question.
More and more people turn to AI when they’re struggling, including people living with abuse. That makes how these tools are designed matter a great deal. A well-designed AI can offer validation, clarity and a sense of control at a time when all three feel out of reach. A poorly designed one can minimize, misread or even mirror the tactics a survivor is trying to escape.
That’s why we were glad to see the Trauma-Informed Algorithmic Assessment Toolkit, published in 2024 by researchers at the ARC Centre of Excellence for Automated Decision-Making and Society at the University of Queensland. It gives organizations a practical way to check whether their AI and automated services are safe for people who have experienced trauma.
We’ve been building trauma-informed AI at Aimee Says since generative AI first reached the public, and the toolkit’s principles match what survivors have told us they need.
The 5 principles of trauma-informed AI
1. Safety
Safety isn’t only about protecting data. It’s also emotional safety: responses that don’t blame, minimize or retraumatize, and that never mimic abusive tactics like gaslighting or control.
2. Trust and transparency
Many survivors have been failed by people and systems they trusted. AI should be upfront about what it can and can’t do, behave predictably, and say clearly when a human professional is needed.
3. Empowerment and choice
Abuse takes away autonomy. AI should give it back: letting people decide what to share, when to engage and what happens to their information, without pressure.
4. Collaboration
Tools for survivors should be built with survivors, advocates and professionals, not just for them.
5. Intersectionality
Race, disability, gender identity, income, immigration status and culture all shape how people experience abuse and seek help. AI should recognize those differences and avoid bias against marginalized communities.
Questions to ask before you trust an AI with your story
Whether you’re a survivor choosing an app or an organization choosing a tool, these questions can help:
A quick checklist
- Is my data used to train AI models? Can I opt out, or is it never used?
- Is my information sold or shared with advertisers?
- Can I delete my data and my account, and how fast does that happen?
- Does it tell me clearly what it is, and what it isn’t (a lawyer, a therapist, an emergency service)?
- Does it ever tell me what I “should” do, or does it leave the decisions to me?
- Were survivors and advocates involved in building it?
- Does it point me to human help, like hotlines, when I need it?
A trauma-informed tool should leave you feeling more in control, not less.
How Aimee puts these principles into practice
- Your data isn’t used for training or sold. No AI is trained on your chats or content, and we never sell your information.
- You stay in control. You decide what to share and when. You can start without an account, and you can delete your data and account at any time.
- Clear about what she is. Aimee is an AI, not a lawyer or therapist, and she says when something needs a human professional.
- No judgment, no instructions. Aimee reflects back what she hears and helps you name patterns. The choices stay yours.
- Built with the field. We work with domestic violence advocates and survivors to shape how Aimee responds, and we change things when survivors tell us what they need.
Read more on our privacy page.
Frequently asked questions
What does “trauma-informed AI” mean?
It means an AI system is designed with an understanding of how trauma affects people, so it prioritizes safety, trust, choice, collaboration and respect for differences, and avoids causing further harm.
Is it safe to talk to a general-purpose chatbot about abuse?
It depends on the tool. Check whether your conversations are used for training, who can see them, and whether you can delete them. If someone may check your devices, think about device safety too.
Who created the Trauma-Informed Algorithmic Assessment Toolkit?
Researchers Suvradip Maitra, Lyndal Sleep, Suzanna Fay and Paul Henman at the ARC Centre of Excellence for Automated Decision-Making and Society, University of Queensland, in 2024.
See it for yourself
Aimee is a private, trauma-informed companion for people facing abuse and coercive control.
Talk to Aimee Free to start. No account needed.