OpenAI's teen ChatGPT is a compliance product, and rivals will be measured against it

OpenAI is launching ChatGPT for Teens on Tuesday, a version of the chatbot built for users aged 13 to 17 with content restrictions covering suicide, self-harm and romantic or sexual conversation, plus homework support designed to guide rather than answer ^[1][2][17]. The interesting part is not the safety marketing. It is that OpenAI has now published a specific list of prohibited model behaviours, and anyone selling a chatbot to a school district or a nervous parent will be asked why their list is shorter.

The context is unflattering. According to Fortune, research from a watchdog group last year found ChatGPT would tell 13-year-olds how to get drunk and high, instruct them on concealing eating disorders, and compose a suicide letter to their parents on request, typically after warning against the activity first ^[5]. A 2025 Common Sense Media study found more than 70% of US teens turning to AI chatbots for companionship, with half using AI companions regularly ^[3]. Sam Altman has described emotional overreliance on the technology as a "really common thing" among young people ^[4]. Shipping a restricted tier before a regulator or a lawsuit specifies one is the cheaper sequence.

The design choices are worth reading closely because they are the parts competitors will have to copy. OpenAI does not verify ages; it estimates whether a user is under 18 from signals such as the kinds of queries they type, and anyone identified or self-identified as a minor is placed into the teen version automatically ^[8][9][10]. Fortune compares this to Meta's teen accounts on Instagram ^[11]. So tier assignment rests on classifier inference rather than documents ^[21], which means the failure modes run in both directions and neither OpenAI's accuracy rate nor its appeal process is disclosed in the source material.

Parental controls require both the teen and the guardian to opt in, and include quiet hours that block access plus safety notifications in limited high-risk situations such as possible self-harm ^[12][13][14]. OpenAI says it is adding notifications related to eating disorders while limiting what is shared ^[15]. Ann O'Leary, OpenAI's vice president of global policy, said the intent is to "make sure that this is safe, even if you don't use parental controls" ^[16]. That is the load-bearing claim: the protections are defaults, not settings, which is the only version that survives contact with teenagers who do not link their accounts.

On companionship, the model is barred from suggesting it has personal feelings toward the user or implying it is conscious or experiences emotions ^[6]. Allison Mishkin, OpenAI's head of child development, said the team identified the hypothetical cues a model could give that might make a teenager form a relationship with it ^[7]. Set against the Common Sense Media figure, OpenAI is deliberately degrading the most popular teen use case for its own product ^[3][6].

The commercial logic points at schools. OpenAI already sells a version for teachers, and Fortune notes that a study-tuned model gives it more routes into education buyers ^[18][19]; Mishkin cited research that people learn more effectively when they actively struggle with concepts ^[20].

Watch three things: whether OpenAI publishes accuracy numbers for age assurance, whether Google, Meta and Character-style companion apps match the ban on simulated feelings or quietly do not, and whether district procurement documents start naming these restrictions as requirements. The last one is how a product decision becomes a standard.