OpenAI is giving ChatGPT a new safety role that goes beyond conversation. With the rollout of Trusted Contact, the company is trying to make the chatbot a faster bridge to human help when a discussion appears to be heading toward self-harm or a mental health crisis.
The feature became available globally on 7 May and is aimed at moments when the system detects serious danger. In those cases, ChatGPT can steer the user toward a crisis hotline or emergency services, while also offering a way to reach a person the user already trusts.
How Trusted Contact works
Trusted Contact lets users choose one trusted adult from within ChatGPT settings. That contact must be over 18, or 19 in South Korea, and does not need to have a ChatGPT account.
If the trusted contact does have an account, the notification process becomes easier. In some situations, ChatGPT may also first ask the user to contact that person when the system detects serious self-harm risk.
OpenAI is also preparing a form of proactive notification to the trusted contact after review by a small team. All safety notifications still go through trained human review before they are sent, and the company says that review is targeted to finish in less than an hour.
Why the feature matters now
The launch comes as chatbot behavior around vulnerable users faces increasing scrutiny. Systems such as ChatGPT and Claude have been criticized in recent years for potentially reinforcing delusional thinking, intensifying paranoia, or raising emotional tension.
That concern has helped push the phrase “AI psychosis” into wider public discussion. The term reflects anxiety that persuasive AI conversations can become harmful when users begin treating the system like a close friend, therapist, or emotional companion.
OpenAI says it understands that pressure. Its safety team has acknowledged that some users have experienced serious mental health crises while interacting with AI systems, even if such cases remain rare.
A safety layer, not a replacement
The company is careful to frame Trusted Contact as an added path to support, not a substitute for professional care or crisis services. The feature is meant to connect users with someone they already trust, which aligns with the idea that social connection can help reduce suicide risk.
OpenAI also notes that no system is perfect. A Trusted Contact notification may not fully capture what a person is experiencing, even after human review.
That tension sits at the center of the new rollout. Stronger monitoring may improve protection, but it can also encourage some users to look for ways around detection.
Still, the direction is clear: OpenAI is pushing ChatGPT closer to a model where the chatbot does more than respond. It is also expected to recognize when a conversation should be handed off to real human support.
Source: www.notebookcheck.net
