New Under-18 principles define how ChatGPT should interact with teenage users.
OpenAI updated its Model Specification to include protections that guide how its AI models interact with adolescent users aged 13 to 17 under its new Under-18 (U18) Principles framework. The Model Spec is the company’s internal set of rules, values, and behavioral expectations for its AI systems, including ChatGPT.
The U18 Principles are intended to shape ChatGPT’s behavior with teen users by prioritizing safety and age-appropriate support. They establish commitments to “put teen safety first,” encourage real-world support and offline relationships, treat teens according to their developmental stage, and be transparent about expectations.
According to OpenAI, the updated guidelines are grounded in developmental science and were shaped with input from external experts, including the American Psychological Association (APA). In a quoted statement, Dr. Arthur C. Evans Jr., CEO of the American Psychological Association, said the association “encourages AI developers to offer developmentally appropriate precautions for youth users” and stressed the importance of supervised and discussed use of AI with trusted adults.
The update clarifies how safety considerations apply when “higher-risk areas” arise in conversations with teens. OpenAI says the assistant should apply stronger guardrails and offer safer alternatives, and when there is imminent risk, it should urge teens to contact emergency services or crisis resources.
The company said the Model Spec’s teen protections build on other efforts, including parental controls that allow adults to tailor ChatGPT experiences for teens, and new expert-vetted resources for families. It also noted that it is in the early stages of rolling out an age prediction system on ChatGPT’s consumer plans that would automatically apply the U18 experience when an account is believed to belong to a minor. If age cannot be confidently determined, OpenAI said it would default to the U18 experience and provide adults with a way to verify their age.
The updated Model Spec and U18 Principles remain part of OpenAI’s intended model behavior and will continue to be refined.