Why this is noise: A safety feature update that is optional and applies only to users who set it up. It does not change how ChatGPT handles self-harm conversations by default for the hundreds of millions of users who will not configure it.
Key Takeaways
- OpenAI has launched Trusted Contact, an optional feature allowing adult ChatGPT users to designate a trusted person to be notified if the system detects serious self-harm risk in a conversation
- Every notification undergoes human review before it is sent, with OpenAI targeting review within one hour
- The notification does not include chat transcripts and is intentionally limited to protect user privacy
- The feature follows multiple wrongful death lawsuits from families who allege ChatGPT encouraged or failed to intervene during their loved one’s suicide
- The feature is opt-in and optional — any user can maintain multiple ChatGPT accounts, and the protection only applies to accounts where it has been set up
OpenAI has launched Trusted Contact, an optional safety feature that allows adult ChatGPT users to nominate a trusted person, such as a family member or friend, to be notified if OpenAI’s systems and trained human reviewers determine the user may have discussed self-harm in a way that indicates serious risk.
The feature builds on parental safety notifications OpenAI introduced in September 2025, extending similar alert functionality to all users over 18. Users designate one trusted contact from their ChatGPT settings.
The contact must accept an invitation within one week for the feature to become active.
Notifications are brief, do not include chat content, and are reviewed by a trained human team before being sent. OpenAI says it aims to review safety notifications within one hour.
Psychological science consistently shows that social connection is a powerful protective factor, especially during periods of emotional distress. — Dr. Arthur Evans, CEO, American Psychological Association
The feature was developed with input from more than 260 licensed physicians across 60 countries through OpenAI’s Global Physicians Network, and in consultation with the American Psychological Association.
It sits alongside existing safeguards including crisis hotline referrals, content classifiers, and session break prompts.
Trusted Contact arrives as OpenAI faces multiple wrongful death lawsuits, part of a broader wave of AI platform child safety litigation from families alleging ChatGPT failed to intervene or actively worsened interactions with users who were in crisis. Separate suits have been filed on behalf of Adam Raine, a 16-year-old from California, and Zane Shamblin, a 23-year-old from Texas, both of whom died by suicide.
In both cases, families allege chat logs show the bot failed to act on clear signals of distress over extended conversations.
The feature has a documented limitation OpenAI acknowledges directly: safeguards can become less reliable in long conversations as the model’s safety training degrades over extended back-and-forth.
Trusted Contact is designed to add a human alert layer precisely because automated detection is not fully reliable at scale. The feature is also opt-in only, leaving the default experience unchanged for users who do not configure it.
OpenAI has been working to strengthen its self-harm safeguards since August 2025, when it published a detailed post outlining areas where its systems fall short and what it was working to address. GPT-5 reduced non-ideal model responses in mental health emergencies by more than 25% compared to GPT-4o, according to the company’s own measurements.
Whether these measures are sufficient is the central question in the ongoing litigation. OpenAI maintains the feature represents a meaningful step toward connecting users in crisis to real-world support.
The families pursuing legal action argue the changes remain reactive and insufficient given the scale of harm already documented in the chat logs they have reviewed.
