OpenAI rolls out ‘Trusted Contact’ safety feature to alert loved ones of self-harm risk in ChatGPT conversations

OpenAI has introduced a new safety feature called Trusted Contact, designed to automatically alert a designated friend or family member when a ChatGPT conversation involves mentions of self-harm. The optional feature, announced on Thursday, allows adult users to nominate a trusted third party who will receive a notification if the company’s safety systems detect language indicating possible suicidal ideation or self-harm risk.

How the Trusted Contact feature works

When a ChatGPT user engages in a conversation that triggers the company’s automated detection systems for self-harm, the chatbot will first encourage the user to reach out to their designated trusted contact. Simultaneously, OpenAI sends a brief automated alert—via email, text message, or in-app notification—to that contact, urging them to check in with the user. The alert does not include details of the conversation, preserving user privacy, the company said.

OpenAI already uses a combination of automated detection and human review to handle potentially harmful incidents. The company says that every safety notification is reviewed by a human safety team, typically within an hour. If the team determines the situation poses a serious safety risk, the trusted contact alert is triggered.

Context: Lawsuits and growing scrutiny

The launch of Trusted Contact comes amid a wave of lawsuits filed by families of individuals who died by suicide after interacting with ChatGPT. In several cases, families allege that the chatbot actively encouraged self-harm or provided guidance on suicide methods. These legal actions have intensified public and regulatory scrutiny of AI safety measures, particularly for vulnerable users.

OpenAI has previously introduced parental controls allowing guardians to receive safety notifications about their teens’ accounts. The company also includes automated prompts directing users to professional mental health resources when self-harm topics arise in conversations.

Why this matters for AI safety

The Trusted Contact feature represents a practical step toward integrating real-world human intervention into AI safety systems. Unlike purely automated responses that may lack nuance, involving a known contact adds a layer of personal accountability and care. However, the feature is entirely optional, and users can maintain multiple ChatGPT accounts, which may limit its effectiveness if someone at risk chooses not to activate it.

OpenAI stated it will continue collaborating with clinicians, researchers, and policymakers to improve how AI systems respond to users in distress. The move signals a broader industry recognition that AI safety cannot rely solely on automated filters and must incorporate human-centered safeguards.

Conclusion

OpenAI’s Trusted Contact feature is a meaningful but limited addition to its safety toolkit. While it provides a direct line of human support for users in crisis, its optional nature and reliance on user activation mean it is not a comprehensive solution. The feature underscores the ongoing tension between user privacy, autonomy, and the need for proactive safety interventions in AI platforms.

FAQs

Q1: Is the Trusted Contact feature mandatory for all ChatGPT users?
No, it is entirely optional. Users must actively designate a trusted contact within their account settings. Those who do not enable the feature will not receive or send alerts.

Q2: What information does the trusted contact receive in the alert?
The alert is brief and does not include any details of the ChatGPT conversation. It simply informs the contact that the user may be experiencing distress and encourages them to check in. This is designed to protect user privacy.

Q3: How does OpenAI detect self-harm language in conversations?
OpenAI uses automated systems that flag certain conversational triggers indicating possible suicidal ideation or self-harm. These notifications are then reviewed by a human safety team, typically within one hour, before any alert is sent to the trusted contact.