OpenAI updates model rules to strengthen protections for teens using ChatGPT
OpenAI has introduced new safeguards for teenagers using ChatGPT, updating its Model Spec to include specific under-18 principles designed to strengthen safety, transparency, and parental oversight.
The changes reflect growing scrutiny around how generative AI tools are used by students aged 13 to 17, and signal a broader shift toward age-appropriate AI design as schools, families, and policymakers grapple with the role of AI in education and everyday life.
New under-18 principles added to Model Spec
In a LinkedIn post, Chris Lehane, chief global affairs officer at OpenAI, says the update is driven by a shared concern among parents, educators, and policymakers about access to “safe, trustworthy AI at home and at school.” He describes the Model Spec as “the instruction manual and rulebook for an AI model,” which “explains how the model is supposed to behave.”
The updated Model Spec introduces a dedicated set of under-18 principles for users aged 13 to 17, structured around four commitments: putting teen safety first, promoting real-world support, treating teens as teens, and setting clear expectations through transparency.
Lehane says the principles provide a roadmap for how OpenAI approaches teen safety, particularly as today’s teens become the first generation to grow up with AI embedded in learning, communication, and early workforce preparation.
The update also clarifies how ChatGPT should respond to higher-risk situations involving teens, including self-harm and graphic content. In those cases, the model is expected to apply stronger guardrails, offer safer alternatives, and encourage users to seek trusted offline support. Where there is imminent risk, teens are directed toward emergency services or crisis resources.
Age-based safeguards and parental controls
Alongside the policy update, OpenAI is expanding how teen safeguards are applied in practice. Lehane says OpenAI is in the early stages of rolling out age prediction to help automatically apply under-18 protections when an account is believed to belong to a minor. Where age is unclear, the system defaults to an under-18 experience, with adults able to verify their age to access full capabilities.
The company has also extended teen safety features across newer products, including group chats, the ChatGPT Atlas browser, and the Sora app. Existing parental controls allow families to manage settings across ChatGPT and Sora, set quiet hours, adjust content controls, and receive safety alerts in rare cases involving acute distress.
Lehane says OpenAI is also exploring how to advance teen safety through engagement at state and federal levels, positioning the updated Model Spec as a foundation for broader policy discussions.
New AI literacy resources for families
In parallel with the Model Spec update, OpenAI has released two new AI literacy resources aimed at teens and parents. One is a family-friendly guide explaining how AI models are trained, why they can make mistakes, and how teens can use tools like ChatGPT more thoughtfully. The guide includes practical advice on writing prompts, double-checking information, managing data, and understanding AI limitations.
The second resource offers conversation starters and guidance for parents, focusing on building critical thinking, setting boundaries, and navigating sensitive or emotional topics. Both resources were developed with input from OpenAI’s Expert Council on Well-Being and AI, alongside external partners including ConnectSafely.
Lehane describes the initiative as part of a longer-term commitment, saying, “This is the first generation growing up with AI. Our job is to make sure safety keeps pace with how these tools are actually used.”
