OpenAI Adds Trusted Contact Alerts to ChatGPT for Self-Harm Safety

OpenAI has introduced Trusted Contact in ChatGPT, an optional safety feature that notifies someone a user trusts if the system detects serious self-harm concerns.

The company described the feature in an announcement titled "Introducing Trusted Contact in ChatGPT." According to OpenAI, the feature is opt-in by design: users choose to activate it and designate a person they trust to receive notifications when ChatGPT's safety systems flag serious concerns about self-harm in a conversation.

The feature addresses a growing tension in AI deployment. As hundreds of millions of people use ChatGPT for open-ended conversation, the system sometimes encounters users in acute distress. How an AI product should respond in those moments — and whether, or when, to involve another human — has become a concrete product and policy question for every major AI developer. OpenAI's answer is to let users decide in advance who should be looped in, rather than making that call unilaterally.

OpenAI frames Trusted Contact as a safety mechanism, not a monitoring tool. The feature only activates when the user has explicitly enabled it, and it only triggers notifications when serious self-harm concerns are detected — not for general conversations or routine safety signals.

The design points to a broader industry shift. AI companies are under increasing pressure from regulators, researchers, and safety auditors to demonstrate that their products handle crisis situations responsibly, particularly when vulnerable users, including minors, treat chatbots as confidants. Opt-in human-notification systems are one of the more direct responses available: they connect automated detection of distress to an actual person who can intervene, while preserving user consent as the foundation.

For OpenAI, the rollout extends the safety infrastructure the company has already built around crisis responses in ChatGPT, which has previously surfaced helpline information in situations involving self-harm risk. Trusted Contact moves beyond surfacing resources to actively enlisting a designated human, closing the gap between detection and intervention.

The stakes are concrete. Detection without follow-up leaves the hardest cases unresolved; automated systems can flag risk but cannot themselves provide sustained human support. By notifying a trusted person, OpenAI shifts part of that responsibility from the model to the user's own chosen network, with the user's prior consent as the basis for that transfer.

OpenAI has not detailed in the announcement how notifications are delivered, what thresholds trigger them, or how the feature handles situations involving minors. Those implementation details will determine whether Trusted Contact works as intended in practice — and whether users, particularly those most at risk, opt in at all.