OpenAI Adds Under-18 Principles to Its Model Spec

OpenAI has updated its Model Spec with new Under-18 Principles that define how ChatGPT should support teens with safe, age-appropriate guidance grounded in developmental science.

The company announced the change as a strengthening of existing guardrails rather than a fresh policy framework. According to OpenAI, the update does three things: it tightens safety protections for younger users, it clarifies how the model is expected to behave in higher-risk situations involving teens, and it extends the company's broader work on teen safety across ChatGPT.

The stakes are straightforward. ChatGPT is one of the most widely used AI products in the world, and teenagers make up a significant share of its user base — a population regulators, parents and researchers have repeatedly flagged as needing distinct treatment from adults. OpenAI's Model Spec functions as the rulebook its models are trained and evaluated against, so changes to it translate into changes in observable product behavior, not just policy language.

What are the Under-18 Principles?

The Under-18 Principles are a new addition to the Model Spec, OpenAI's published specification for how ChatGPT ought to behave. In its announcement, OpenAI said the principles define how ChatGPT should support teens with "safe, age-appropriate guidance grounded in developmental science."

That framing matters. It signals that OpenAI is grounding its teen-facing behavior rules in research on adolescent development rather than treating minors as a single undifferentiated user category. The company said the update "strengthens guardrails" — its own words — for how the model engages with users under 18.

The second stated goal is behavioral clarity in edge cases. OpenAI said the update "clarifies expected model behavior in higher-risk situations." The company did not enumerate those situations in its announcement, but the phrase points to the category of interactions — topics touching safety, wellbeing and vulnerability — where the cost of a wrong answer from a chatbot is highest.

Third, OpenAI positioned the change as continuous with, not separate from, its existing safety work. The update "builds on our broader work to improve teen safety across ChatGPT," the company said.

Why does the Model Spec matter?

The Model Spec is OpenAI's attempt to write down, in public, the values and behavioral rules its models should follow. It covers how the models should handle sensitive topics, how they should respond to different user contexts, and where they should refuse or escalate. Because OpenAI uses the Spec in training and evaluation, revisions to it are among the most direct levers the company has for changing what hundreds of millions of users actually experience.

Adding age-specific principles to that document elevates teen safety from a product-team concern to a specification-level commitment. That distinction carries weight for anyone auditing the company's claims: behavior written into the Spec is behavior OpenAI has declared itself accountable for shipping.

It also matters in the current regulatory climate. Lawmakers in the United States and Europe have pressed AI companies on minors' protections, and age-appropriate design has become a baseline expectation for consumer platforms, not a differentiator. A published, science-grounded set of under-18 principles gives OpenAI something concrete to point to in those conversations.

What changes for teens using ChatGPT?

Based on OpenAI's announcement, the practical effect is a tightened and clarified set of behaviors:

• Age-appropriate guidance. ChatGPT's responses to teen users should reflect what OpenAI calls guidance "grounded in developmental science." The company said the principles define "how ChatGPT should support teens."
• Stronger guardrails. OpenAI states plainly that the update "strengthens guardrails" for younger users, implying the previous Spec offered less explicit direction on under-18 interactions.
• Clearer behavior in high-risk moments. The Spec now spells out expected model behavior "in higher-risk situations," reducing ambiguity about what the model should do when a conversation touches on danger or vulnerability.

OpenAI did not publish specific behavioral examples, benchmark figures or enforcement mechanisms alongside the announcement, so the depth of the change will become visible only as users and researchers probe the updated model behavior in practice.

What comes next?

OpenAI frames this as one step in ongoing work. The company said the Under-18 Principles build on its "broader work to improve teen safety across ChatGPT," language that suggests further refinements — in the Spec, in product features, or both — are likely to follow as the company tests how the new principles hold up against real-world use by teenage users.