Updating our Model Spec with teen protections
OpenAI ● Covered by 2 sources
OpenAI is rewriting ChatGPT's rulebook for users under 18, adding specific guardrails for teens. It's the clearest sign yet that regulators and lawsuits are pushing AI companies to treat minors differently.
Based on reporting by OpenAI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI just added a new chapter to its Model Spec, the internal document that governs how ChatGPT is supposed to behave, and this one is entirely about teenagers. The company calls them Under-18 Principles, and the idea is to ground the model's responses to younger users in actual developmental science rather than a one-size-fits-all approach to safety.
What's changing in practice is the handling of higher-risk situations. Think conversations that touch on self-harm, eating disorders, substance use, or other topics where a generic chatbot answer could do real damage to a 15-year-old in a way it wouldn't to a 35-year-old. OpenAI says the update clarifies exactly what the model should and shouldn't say when it detects it might be talking to a minor, tightening guardrails that were previously looser or more ambiguous.
This isn't happening in a vacuum. OpenAI has been under pressure for months over teen safety, including scrutiny tied to reports of ChatGPT conversations preceding self-harm incidents involving young users. The company frames this Model Spec update as part of a broader, ongoing effort rather than a one-off fix, which suggests more changes to age-detection and content filtering are likely coming.
The harder problem, and one OpenAI doesn't fully address here, is how ChatGPT actually knows it's talking to someone under 18 in the first place. Age verification online has always been a mess, and a chatbot has even less to go on than a website with a birthdate field. Writing better rules for teens is one thing; reliably knowing who's on the other end of the conversation is a much messier engineering and policy problem.
My take — AI-written commentary, not fact-checked reporting
I think this is OpenAI doing the right thing a little late, which describes basically every AI safety move of the past two years. Writing careful rules for teens is good, but until there's a real way to verify age, these principles are mostly aspirational text sitting on top of a system that still can't tell a 14-year-old from an adult. Watch for regulators to force the harder part soon.
Read more about this at: OpenAI
