ChatGPTに緊急時対応の信頼できる連絡先機能を導入 Introducing Trusted Contact in ChatGPT

OpenAI has added a feature called Trusted Contact to ChatGPT, an optional safety mechanism that can notify a person a user has designated in advance when the system detects serious concerns about self-harm. The addition matters because conversational AI assistants are increasingly used in moments of emotional distress, and the question of how such systems should respond to potential crises has become one of the most scrutinized aspects of consumer AI.

The way the feature works is straightforward in concept. A user can choose to register one or more people they trust, such as a family member or close friend, ahead of time. If ChatGPT subsequently detects signals that suggest a serious risk of self-harm during a conversation, it can reach out to that designated contact. The intent, as described by OpenAI, is to create a faster path to human support during a crisis, connecting an at-risk person with someone who knows them rather than leaving them alone with an automated tool. Because the feature is opt-in, users decide whether to enable it and whom to list.

Several details shape how the feature is likely to be received. First, it is positioned as an additional layer rather than a replacement for existing safeguards. ChatGPT already responds to expressions of distress by surfacing crisis hotline information and encouraging users to seek professional help in many regions. Trusted Contact appears designed to address a gap in that approach, namely that pointing someone toward a hotline does not guarantee they will act on it, whereas alerting a known person introduces accountability and personal connection. Second, the reliability of any such system depends heavily on detection accuracy. Automated assessment of whether a conversation reflects a genuine emergency is difficult, and both false positives, which could breach a user's privacy or trust, and false negatives, which could miss a real crisis, are meaningful risks. OpenAI has not, in the available summary, detailed the precise thresholds or escalation logic, and these operational specifics will likely determine how well the feature performs in practice.

The move sits within a broader industry context. Over the past few years, AI developers have faced growing pressure from regulators, mental health professionals, and the public over how chatbots handle vulnerable users. Reports of people forming intense attachments to AI companions, along with concerns that some systems have responded poorly to discussions of suicide or self-harm, have intensified calls for stronger trust and safety practices. Companies including Anthropic, Google, and Meta have all invested in content moderation and crisis-response behaviors for their conversational products, and several jurisdictions are weighing rules that would require clearer safeguards for minors and at-risk users. Trusted Contact can be read as part of this wider effort to demonstrate responsibility as these tools reach hundreds of millions of people.

Privacy and consent are central tensions that the design must navigate. Notifying a third party about a user's mental state is sensitive by nature, and the optional, pre-registered structure appears intended to keep the user in control of when and how their information is shared. Questions that remain relevant include what exactly a trusted contact is told, whether the user is informed when a notification is sent, and how the data involved is stored and protected. How OpenAI communicates these boundaries will influence whether users feel the feature is supportive rather than surveilling.

It is also worth noting the categorization of this announcement under Codex, which is associated with OpenAI's developer and engineering work. This may indicate that the feature involves underlying tooling or infrastructure changes, though the user-facing description emphasizes safety rather than developer functionality. As with many incremental safety updates, the rollout details, including which markets and which versions of ChatGPT receive the feature first, will shape its real-world impact.

For users and observers, the practical takeaway is that ChatGPT is continuing to expand the human element of its crisis response rather than relying solely on automated messaging. Whether Trusted Contact meaningfully improves outcomes will depend on adoption rates, the accuracy of its detection systems, and the willingness of designated contacts to respond. The feature represents a measured step in an area where the stakes are high and where the appropriate role of AI remains actively debated among clinicians, ethicists, and technologists.