Concern 582 · Working draft · editorial review in progress
Chatbots producing self-harm content
Unsafe self-harm responses occur when a chatbot encourages, enables or fails to respond appropriately to a person expressing a wish to hurt themselves.
Understanding the concern
The immediate concern is the content and direction of the reply. A system may sound sympathetic while missing the seriousness of the situation or reinforcing harmful thinking. General fluency does not establish the ability to recognize and respond safely to distress.
Safety evaluations have identified failures in tested systems. Those findings support careful safeguards, but they should not be used to claim that a chatbot alone explains any individual's actions. A responsible service should avoid harmful assistance and help connect the person with appropriate human support, with clear limits on what the service can do.
An example
A distressed user receives an agreeable reply when the situation calls for a careful, supportive safety response.
Illustrative unless identified as a documented case.
Evidence and its limits
These studies test responses to scenarios. They identify safety failures without establishing a population rate of self-harm caused by chatbots.
Moore and colleagues · FAccT 2025 · Safety evaluation
Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers ↗ (opens in a new tab)Tests stigma and responses to clinical scenarios; identifies failures without estimating population rates of injury.
Brewster and colleagues; JAMA Network Open · Research study
Characteristics and Safety of Consumer Chatbots for Emergent Adolescent Health Concerns ↗ (opens in a new tab)2025 simulated conversations and expert-rated responses, not a clinical trial measuring resulting injuries.
Broader frameworks and background
These resources provide context. They do not establish every claim in this concern.
A question to discuss
How should a chatbot recognize that a conversation requires human support rather than another ordinary answer?
Keep track of different views, what needs checking, and any next step in your own discussion notes.
Open discussion notes →Principles for the conversation
Before you finish · reflect together
Virtues and principles to consider
Virtues are qualities we practice in how we treat people and make choices. They are not scores for judging someone. Use these ideas to explore what a thoughtful response could look like.
Dignity
Dignity means recognizing each person's worth regardless of usefulness, status, or a system's score. People deserve respect and a voice in matters that affect them.
Explain important decisions, avoid humiliating treatment, and keep room for a person's story. Do not treat a prediction as a complete account of who someone is.
Ask together: Does this use treat each person as someone whose experience matters?
Listening
Listening means trying to understand someone's experience before preparing an answer. People who use a service, work with it, or are judged by it may notice problems its designers miss.
Invite people to describe what happened in their own words. Repeat back what you understood and let them correct it. A quiet person or a complaint that is hard to measure may still reveal something important.
Ask together: Whose experience have we not heard yet?
Seeing potential
Seeing potential means recognizing people's capacity to learn, contribute, and grow. It resists reducing someone to a score or to the tasks a machine can perform faster.
Use tools to support people's development. Give learners time, feedback, and real opportunities to practice. A prediction about someone's performance should not close off their future.
Ask together: How could this help people develop abilities they value?
Editorial ratings
These are editorial judgments from the register, not measured probabilities or predictions about your situation.
About the register and scoring →More ways to explore
Search for more writing
For more specific or recent material, open a search below or ask for a web summary. AI-generated summaries can contain mistakes; open the original articles to check them.