OpenAI didn’t build ChatGPT for Teens because it wanted to. It built it because it had to. Last September, after the company was sued for allegedly contributing to the death of 16-year-old Adam Raine, OpenAI announced it was working on a system to automatically detect and restrict teen users. That system is now live. And while it’s easy to be cynical about corporate safety initiatives that arrive with legal pressure attached, what OpenAI has shipped here is more substantive than a simple content filter.
The core promise of ChatGPT for Teens is automatic enrollment. If OpenAI’s system predicts a user is under 18, or if the user has self-reported their age, the new experience becomes the default without any action required. Lauren Jonas, OpenAI’s head of youth and families, put it plainly: “There’s no need for a teen to create a new account or change anything.” That’s the right approach. Systems that require teens to opt into safer modes are systems teens will ignore.
The experience is built around two areas: learning and safety. On the learning side, OpenAI is leaning hard into Study Mode, a Socratic-style interaction model that pushes back on direct answers and instead guides users toward working through problems themselves. It won’t be the default setting, but OpenAI is introducing Study Hours, a feature that automatically activates Study Mode during set times. Parents with linked accounts can configure this, or teens can set it themselves. The system will also detect when a young user appears to be trying to get ChatGPT to do their homework and redirect them accordingly.
The visual explanation feature inside Study Mode has also expanded significantly. When it launched in March, ChatGPT could generate educational visuals across roughly 70 topics. That number is now above 300, covering subjects like integrals, mitosis, and moon phases. OpenAI says it has run randomized controlled trials on Study Mode and is seeing positive performance gains, though Jonas declined to share specific numbers.
The safety side of the update is where things get more interesting from an industry perspective. OpenAI is explicitly restricting the model from positioning itself as a friend, expressing personal feelings, or implying sentience when talking to teen users. This is a direct response to concerns that have followed Character.AI and, to a lesser degree, ChatGPT itself: that conversational AI can blur into something that feels like a relationship, particularly for younger users who may be more vulnerable to that dynamic. Break reminders will now appear more frequently and will explicitly frame ChatGPT as a tool, not a companion. Quiet hours, where ChatGPT becomes unavailable entirely, can also be set by teens or their parents.
OpenAI is also expanding its parental notification system. Last month it began alerting parents with linked accounts when a teen was removed for violating policies around violent threats. Now that system includes conversations flagged for eating disorder-related concerns. According to Jonas, any conversation indicating “serious self-harm concerns” can trigger an alert, with review by a full-time OpenAI employee before a parent is notified, and a target response time of under one hour.
The piece that remains genuinely unclear is how reliable the age detection actually is. Discord, Snapchat, and other platforms have all implemented age verification or detection systems in recent years, and teens have consistently found ways around them. OpenAI hasn’t shared accuracy data for its prediction model, and that gap matters. A system that catches 70 percent of underage users still leaves a meaningful number unprotected. Jonas acknowledged that this is a starting point, not a finished product. Given the regulatory and reputational pressure the company is under, it had better treat it that way.
ChatGPT for Teens is rolling out globally starting today, with full rollout expected within two weeks.
