OpenAI Launches ChatGPT for Teens With Safety Guardrails

OpenAI is rolling out a dedicated ChatGPT experience for teenagers, adding safety guardrails and learning tools to the world's most widely used consumer AI chatbot. The new ChatGPT for Teens automatically applies to users who state they are between 13 and 17, or when OpenAI's age-estimation system flags an account as under 18, according to OpenAI.

What changed

ChatGPT for Teens blocks romantic and sexual roleplay, refuses to imply it has feelings or consciousness, and reduces exposure to content around self-harm, violence, eating disorders, and dangerous activities. The safeguards draw on an Under-18 Principles section OpenAI added to its Model Spec, which the company says was shaped by developmental science and expert guidance.

On the learning side, a new Study Mode replaces simple answer-giving with guiding questions and step-by-step support. Homework reminders redirect teens back to Study Mode when the model detects they are angling for a direct answer rather than trying to understand the material, TechCrunch reports.

Parental controls

Parents can link their account to a teen's account to manage settings, receive safety notifications, and set Quiet Hours. Extra controls let parents turn off voice mode, disable memory, remove image generation, and opt a teen's conversations out of model training. In high-risk situations, OpenAI says a small team of trained reviewers will notify parents by email, text, and push alert.

Why it matters

The launch lands amid a string of lawsuits over AI chatbots' safety measures and mounting concern about teen mental health. OpenAI says ChatGPT scaled to 900 million weekly users before teen-specific safeguards arrived, and the company is partnering with CodeAI to help teens learn how the technology works and how to question it.

For AI builders, the move signals a broader shift: consumer AI platforms are shipping age-gating, safety defaults, and parental governance as first-class product features rather than afterthoughts. The same guardrail patterns apply broadly, from model-level content restrictions and opt-out training controls to moderation escalation, and they are increasingly relevant to anyone building user-facing or agentic AI products. It is a preview of the safety architecture regulators will expect everywhere, echoing warnings that AI safety tests themselves are becoming a security risk.

Rollout begins this week and expands globally over the coming weeks.