OpenAI's teen tier ships the same week 29 states put Meta's teen design on trial

OpenAI released ChatGPT for Teens, which automatically places users who are, or are estimated to be, 13 to 17 into a version with stronger safety protections, reminders about their use of the tool, and additional parental controls ^[2]. It arrived in the same week that 29 attorneys general brought a lawsuit against Meta in a California court, alleging the company knowingly designed Facebook and Instagram to addict children, misled the public about the harms its platforms posed to young users, and illegally collected data from children under 13 without parental consent ^[1].

Read together, the two events reframe what an age tier is. It is no longer a late trust-and-safety line item. It is the artifact a company wants on the record before someone else's discovery process becomes the industry's design review.

The timelines are the useful number. ChatGPT for Teens arrives nearly four years after ChatGPT launched in November 2022 ^[3]. Instagram had been around for 14 years before Meta introduced dedicated Teen Accounts in September 2024, though Meta products carried other child safety features well before that ^[4]. Same sequence, roughly three and a half times faster ^[1].

OpenAI says the teen version builds on measures introduced over the past year, including parental controls, age prediction, and an under-18 version of its Model Spec, and describes its approach as prioritising teen safety and encouraging real-world support ^[5]. Neither Meta nor OpenAI responded to Fast Company's request for comment ^[6].

Victoria Baines, a former Facebook employee and professor emerita of information technology at Gresham College, told Fast Company that she sees similar trajectories in the safety issues of social media and large language models ^[7]. Her description of how that work gets sequenced is the part operators should sit with: companies start with the most egregious harms, such as child abuse material and terrorism, which are obviously illegal and can be detected and prevented at scale using established tools like matching against known-image databases or blocking guardrails ^[8]. That is rational, she says, because it aims safety spend at compliance risks with a basis in established regulation ^[9].

What sits outside that band is harder. Baines points to risks that are not unlawful but that most people would see as clearly problematic, including age-inappropriate content on social platforms and, for language models, undue influence and misinformation ^[10]. Meetali Jain of the Tech Justice Law Project calls generative AI "social media on steroids," with influence reaching into "the very essence of what it means to be human" ^[11]. Nina Vasan, who directs Stanford's Brainstorm lab, puts the difference plainly: with AI the concern is sharper, because a teenager is not just consuming content ^[12].

Two constraints make the fast-follow tier less reassuring than it looks. Baines says the US lawsuits have made it unclear whether the people focused on growth listen to the people focused on safety ^[13], and she says AI's adoption speed leaves developers less time to find and fix harms than social media companies had ^[14]. A tier gated on estimated age also inherits the error rate of the estimator, and OpenAI's own description concedes the gate runs on estimation as well as declaration ^[2].

Watch whether the Meta trial's theory travels: the claim is about design, not individual pieces of content ^[1], and that is the theory an under-18 model spec is built to answer. Watch, too, whether the harder category Baines describes ever gets a product surface, or whether age-tiering stays parked where regulation already is ^{[9] [10]}.