Suicide Prevention Experts Want AI Chatbots Linked to 988

Published July 31, 2026

Suicide prevention researchers are pressing AI companies to connect their chatbots directly to the existing crisis system rather than leaving users to reach out on their own. The push follows reporting on a young woman who disclosed suicidal thoughts to ChatGPT months before her death without telling her therapist, her parents, or her closest friend.

For anyone weighing mental health treatment options, the reporting lands on one point. When comparing AI to real therapists, a general-purpose chatbot can produce reasonable-sounding advice while doing none of the clinical work that makes an assessment useful.

The New Crisis Feature in Action

In May, OpenAI added a feature called trusted contact. A user names someone they trust, and ChatGPT asks whether that person consents to the role. If the system later detects signs of suicide or self-harm risk, it prompts the user to reach out. It also flags the exchange to a team at OpenAI, and if they agree there’s a safety risk, the trusted contact is notified to check in.

Whether it helps isn’t yet known. OpenAI announced it has seen good adoption of the feature but shared no data. However, other companies have started using AI to flag mental health conditions for those at risk from self-harm.

Experts Say Suicide Prevention Falls Short

Holly Wilcox, a suicide prevention expert at the Johns Hopkins School of Public Health, called it a step in the right direction, though a small one. Her objection is structural. The feature puts the burden back on the person in crisis to have set it up beforehand and to act on a prompt at the worst possible moment.

What Wilcox would rather see is AI companies partnering with crisis infrastructure that already exists. That way, a trained 988 counselor could enter a conversation directly when a user begins describing self-harm or suicidal thinking. That would move the response from a notification to an actual clinical contact.

The Report’s Findings

Sophie Rottenberg was 29 when she died by suicide in February 2025. Her mother, Laura Reiley, relayed that Sophie had described sleeplessness and anxiety the previous fall, used the word depression by Thanksgiving, then moved in with her parents in Ithaca, New York, where they helped her find a therapist.

Reiley noted Sophie never revealed the depth of what she was experiencing to anyone, and her therapist didn’t know she was suicidal. Months later, Sophie’s best friend found a long record of chatbot conversations on her laptop. Reiley shared nearly 1,800 pages.

Sophie had set up a ChatGPT persona that introduced itself as her personal therapist. When she described anxiety, the AI offered evidence-based steps. When she disclosed suicidal thoughts, it acknowledged her and pointed her toward a therapist and 988. It initially declined a request to help write a suicide note, but eventually produced the letter she left behind.

Reiley doesn’t believe ChatGPT caused her daughter’s death. Her view is that if Sophie hadn’t had it as a regular outlet, as many young people do with AI, more of what she was carrying might have surfaced in her relationships with people. Reiley now advocates for regulation.

Clinical Assessment is Different

Reiley’s criticism is worth attention from anyone evaluating mental health treatment facilities. A tool calling itself a therapist didn’t do what a therapist does. It ran no risk assessment to gauge how likely she was to act, and it did not push back.

Psychologist Ursula Whiteside reviewed the chat logs and runs the suicide prevention nonprofit Now Matters Now. The chatbot responded to what Sophie asked for, creating a cycle of affirmation, rather than to her state of mind. Whiteside, who has her own history of suicidal experience, emphasized that connection is protective. She described speaking honestly to another person as an act of recovery in itself.

In general, mental health experts believe general-use chatbots have become considerably safer over the past year. But that remains their assessment rather than a measured finding.

Access Gaps and Co-Occurring Conditions

People don’t usually substitute a chatbot for care because they prefer it. They do it because it is free, immediate, and available at 3 a.m.

That gap falls hardest on people managing more than one condition at once. Someone at risk of suicide due to depression or anxiety alongside a substance use disorder needs integrated treatment for both, which is harder to find and slower to start than opening an app. Co-occurring disorders, also called dual diagnosis, respond better when treated together by a coordinated team than separately or not at all.

Mental Health and Behavioral Health Treatment

If you’ve been leaning on a chatbot to manage something serious, that’s worth treating as information rather than as a failure. A few concrete steps:

Search mental health treatment facilities and behavioral treatment centers by location and level of care

Ask directly whether a program conducts formal risk assessment at intake

Look for dual diagnosis programs if a substance use disorder is also in the picture

Ask whether the program offers group and trauma-focused therapy alongside medication management

Tell a human clinician the part you have only told a chatbot

That’s where we can help. Browse our comprehensive directory of mental health treatment facilities, behavioral treatment centers, and dual diagnosis programs. You can also feel free to call 800-908-4823 (Sponsored) to talk through which level of care fits.

Author

Courtney Myers, MS

Read Bio

Courtney Myers has more than 15 years of experience in online writing and editing. Since graduating from N.C. State University with an MS in Technical Communication, she’s helped clients improve their visibility and reach through expert-level content creation. She specializes in addiction recovery and behavioral healthcare topics.

Editor

Peter Lee, PhD