Revising our Model Spec to add safeguards for teenage users

If you're a parent, educator, or anyone curious about how ChatGPT handles younger users, OpenAI has just announced a meaningful change worth knowing about. The company is revising its Model Spec—the internal rulebook that guides how its AI models should behave—by introducing a new set of Under-18 Principles. These principles spell out how ChatGPT should support teenagers with guidance that is both safe and appropriate for their age, drawing on research from developmental science. The update tightens existing guardrails, describes more clearly how the model should respond in higher-risk scenarios, and extends OpenAI's ongoing efforts to make ChatGPT safer for teens.

What the Model Spec Actually Is

The Model Spec might sound like dry technical documentation, but it plays a central role in shaping ChatGPT. Think of it as a written set of expectations that tells the model what kinds of behavior are acceptable, what should be refused, and how it should navigate tricky situations. When OpenAI updates the Model Spec, those changes influence how the assistant responds in real conversations.

The newly added Under-18 Principles focus specifically on teenage users. Rather than treating all users identically, the updated guidance directs the model to consider the needs and vulnerabilities of younger people, offering support in a way that reflects what developmental science says about adolescence.

What the Under-18 Principles Cover

According to OpenAI, the new principles define how ChatGPT should provide safe, age-appropriate help to teens. A few themes stand out:

• Developmental grounding: The guidance is informed by developmental science, meaning the expected behavior accounts for how teenagers think, mature, and handle sensitive topics.
• Strengthened guardrails: The update reinforces existing protections rather than replacing them, layering additional safeguards on top of what was already in place.
• Clearer expectations in risky situations: One of the more practical changes is a clearer description of how the model should behave when conversations move into higher-risk territory. While OpenAI's announcement does not detail every specific scenario, the company says the update "clarifies expected model behavior in higher-risk situations."

Why This Update Matters

AI assistants are now widely used by students and teenagers, which raises legitimate questions about whether a tool built primarily for adults can serve younger users responsibly. Guidelines written only with adults in mind can fall short when a teen asks about something sensitive or emotionally difficult.

By codifying teen-specific principles directly into the Model Spec, OpenAI is making age-appropriate behavior an explicit requirement rather than an afterthought. This kind of documentation also matters for transparency: the Model Spec is public, so researchers, parents, and policymakers can examine the standards the company has committed to.

Part of a Larger Safety Effort

OpenAI frames this change as a continuation of its broader work on teen safety across ChatGPT, not a standalone fix. The announcement describes the update as building on earlier improvements, suggesting that protecting younger users is an ongoing program rather than a one-time patch.

It's worth noting that the announcement is relatively brief and does not include detailed examples of how the principles will play out in specific conversations, nor does it specify enforcement mechanisms or timelines beyond the update itself. Anyone wanting the full picture should consult OpenAI's published Model Spec documentation for the complete text of the principles.

What Happens Next

As the updated Model Spec takes effect, the practical test will be how ChatGPT actually responds when teens bring up difficult or sensitive subjects. Documentation like this sets the standard, but real-world behavior depends on how well models follow it. For now, the update signals that OpenAI is formally committing to developmental-science-informed, age-appropriate treatment of teenage users—and that its safety rules for teens are now written down for everyone to see.