OpenAI Previews 120-Day Push on ChatGPT Crisis Response and Teen Safety

OpenAI says parents will be able to link their accounts to their teenagers' ChatGPT accounts and receive alerts when the system detects acute distress, as part of a 120-day initiative to improve how the chatbot handles sensitive moments.

The company published the plans in a blog post titled "Building more helpful ChatGPT experiences for everyone," previewing work it says has already started but committing to launch as much of it as possible this year. "We want to proactively preview our plans for the next 120 days, so you won't need to wait for launches to see where we're headed," OpenAI wrote. "The work will continue well beyond this period of time, but we're making a focused effort to launch as many of these improvements as possible this year."

The announcement follows last week's publication of four focus areas for what OpenAI calls "helping people when they need it most": expanding interventions to more people in crisis, making it easier to reach emergency services and experts, enabling connections to trusted contacts, and strengthening protections for teens.

The stakes are significant. Millions of people already use ChatGPT for emotional support and in moments of personal crisis, and regulators, parents, and researchers have pressed AI companies on what their products should do when a user appears distressed — particularly when that user is a minor. OpenAI's answer combines expert oversight, model architecture, and new family controls.

A council of experts and a network of 250 physicians

OpenAI says two bodies now guide its work on well-being: a newly formed Expert Council on Well-Being and AI and a Global Physician Network.

The council, convened earlier this year, brings together experts in youth development, mental health, and human-computer interaction. Its role, according to OpenAI, is "to shape a clear, evidence-based vision for how AI can support people's well-being and help them thrive." Council input will help OpenAI define and measure well-being, set priorities, and design future safeguards, including future iterations of parental controls. OpenAI was explicit about the limits of the arrangement: "While the council will advise on our product, research, and policy decisions, OpenAI remains accountable for the choices we make."

The Global Physician Network is larger and more operational. OpenAI describes it as a pool of more than 250 physicians who have practiced in 60 countries, worked with over the past year on efforts including the company's health bench evaluations, which are designed to measure the capabilities of AI systems for health. Of that pool, more than 90 physicians across 30 countries — psychiatrists, pediatricians, and general practitioners among them — have already contributed to research on how OpenAI's models should behave in mental health contexts. Their input directly informs safety research, model training, and other interventions, and lets OpenAI "quickly engage the right specialists when needed."

The company says it is expanding the network with clinicians and researchers specializing in eating disorders, substance use, and adolescent health. OpenAI promises more detail on both bodies during the 120-day initiative.

Routing sensitive conversations to reasoning models

The most technically notable change involves how OpenAI chooses which model answers a given conversation. Its reasoning models — GPT-5-thinking and o3 — spend more time reasoning through context before answering. OpenAI says testing shows these models, trained with a method the company calls deliberative alignment, more consistently follow safety guidelines and are more resistant to adversarial prompts.

OpenAI recently introduced a real-time router, shipped alongside GPT-5, that selects between efficient chat models and reasoning models based on conversation context. Soon, the company will route some sensitive conversations — such as those where the system detects signs of acute distress — to a reasoning model like GPT-5-thinking, regardless of which model the user first selected. "We'll iterate on this approach thoughtfully," the company added.

The move ties safety behavior to infrastructure rather than to policy text alone. If the router works as described, a user who starts a conversation on a faster, cheaper model would get the more careful model exactly when the conversation turns serious — without needing to know anything about model selection.

Parental controls arrive within a month

The concrete product changes target teenagers, whom OpenAI describes as among the first "AI natives" — a generation growing up with these tools the way earlier generations grew up with the internet or smartphones. "That creates real opportunities for support, learning, and creativity, but it also means families and teens may need support in setting healthy guidelines that fit a teen's unique stage of development," the company wrote.

Within the next month, OpenAI says parents will be able to do four things:

• Link their account with their teen's account, with a minimum age of 13, through a simple email invitation.
• Control how ChatGPT responds to their teen through age-appropriate model behavior rules, which will be on by default.
• Manage which features to disable, including memory and chat history.
• Receive notifications when the system detects their teen is in a moment of acute distress.

OpenAI notes that expert input will guide the distress-notification feature, framing it as a way "to support trust between parents and teens" — an acknowledgment that alerting parents to a teenager's private conversations raises its own questions about surveillance and adolescent privacy.

These controls build on features OpenAI has already rolled out to all users, including in-app reminders during long sessions that encourage taking breaks.

The company calls these steps "only the beginning" and says it will keep strengthening its approach under expert guidance, with progress reports promised over the coming 120 days. The near-term test will be execution: whether parental controls ship on schedule within a month, and whether reasoning-model routing measurably changes how ChatGPT behaves when someone needs help most.