OpenAI has introduced a new safety mechanism called ‘Trusted Contact,’ designed to provide a support network for adult ChatGPT users who may be expressing signs of self-harm. The feature allows users to designate a friend or family member within their account settings. If the platform’s monitoring systems detect language indicative of a potential crisis, the system will prompt the user to contact their designated individual. Furthermore, the company will send an automated alert to the trusted contact, encouraging them to reach out and check on the user’s well-being.

This development follows significant scrutiny regarding the role of AI in mental health, including legal challenges from families who have alleged that interactions with the chatbot exacerbated or contributed to tragic outcomes. In response to these concerns, the company has refined its safety protocols, which now integrate a blend of automated triggers and human oversight. According to the company, notifications regarding potential safety risks are prioritized for human review, with an internal goal of assessing these situations in under one hour.

To balance user privacy with safety, the alerts sent to designated contacts are intentionally brief. They are intended to prompt a wellness check without disclosing the specific details or private content of the conversation that triggered the alert. These notifications can be delivered via email, text message, or in-app alerts, depending on the user’s configuration. The feature is entirely optional and serves as an extension of previous safety initiatives, such as parental oversight tools for teen accounts and automated referrals to professional mental health resources.

While the introduction of Trusted Contact marks a shift toward more proactive user intervention, the company acknowledges that these tools are part of an ongoing effort. The organization stated that it continues to collaborate with clinicians, researchers, and policymakers to refine how artificial intelligence responds to users in distress. Despite these advancements, industry observers note that such features remain elective, meaning they are only effective when users choose to engage with the safety infrastructure provided.