OpenAI Launches ChatGPT for Teens and Sets the Standard That the Law Has Yet to Write
Product for ages 13 to 17 uses age prediction instead of verification and anticipates some of the regulation that American states, the European Union, and Brazil are still negotiating, in the week that the Meta trial began in Oakland.
OpenAI began rolling out ChatGPT for Teens on August 18, a version for users aged 13 to 17 available globally in Free and paid personal plans. The company states that the experience is activated automatically when account signals, age indicated, or algorithmic predictions suggest a minor, and that in some countries, official identification may be requested. In OpenAI's own words, the mechanism is 'age assurance', not verification. This distinction is at the heart of the legal maneuver.
The Product
ChatGPT for Teens blocks responses regarding suicide, self-harm, eating disorders, violence, dangerous activities, and explicit sexual content. It replaces the default response to school tasks with Study Mode and 'responsible homework reminders' that recognize when a teen is attempting to skip their homework. Parents receive control over memory, blackout hours, and response rules at the end of the month, according to the schedule released by the company. The technological foundation remains the same model as the adult products; the difference lies in the policy system and a specific set of guardrails trained for this age group.
Timing is crucial. On the same day as the launch, the federal court in Oakland opened the trial of 29 state attorneys general against Meta for minor addiction and illegal data collection of children under 13. OpenAI has faced months of pressure from Utah, Texas, and California, which passed laws between 2024 and 2026 requiring platforms to estimate users' ages (SB194 in Utah, SCOPE Act in Texas, SB976 in California). The product launched today is the compliant response: instead of waiting for each state to define different criteria, the company applies the stricter standard by default worldwide.
Where the Decision Displeases and Why
The controversial point lies in what OpenAI itself admits: prediction instead of checking. The system can err in both directions, and the alternative offered to an adult labeled as a minor is to present official documentation. This irritates two fields simultaneously. For privacy associations like the Electronic Frontier Foundation, the request for ID normalizes the collection of sensitive data in everyday services. For child mental health groups, the algorithmic prediction is insufficient because teens actively circumvent detectors.
Meta and TikTok have already attempted versions of this compromise and have faced criticism on both ends. Here the difference is editorial: OpenAI is a horizontal infrastructure present in corporate productivity, education, and scientific research, and not a social network. If Study Mode gains traction within American schools already using ChatGPT Edu, such as Houston ISD, Fairfax County, and Fulton County in Georgia, the product for minors ceases to be a legal defense line and becomes a gateway for large-scale educational licensing contracts. This is the market that was needed to justify large-scale AI deployment in American public schools.
What Changes in the Rest of the World
The United Kingdom has, from the turn of 2025 to 2026, the requirement of the Online Safety Act for 'highly effective age assurance' on platforms hosting adult or risky content. Ofcom has already indicated that generative AI services fall within the scope. The approach OpenAI adopts now fits the letter of the British regulation, but the test will come with the first fines, calibrated to up to 10% of global revenue.
In Germany and the rest of the European Union, the new product encounters the Digital Services Act and the Jugendmedienschutz-Staatsvertrag, which obliges platforms to protect minors under 14 with different standards from those applied to older teens. OpenAI's binary model of 'under 18' still does not resolve this distinction in age brackets imposed by German legislation, and the BfDI, the federal data protection authority, is expected to seek formal clarifications in the coming weeks.
In Brazil, the Statute of Children and Adolescents combined with Article 14 of the LGPD already requires specific parental consent for the processing of data for minors under 12. The algorithmic age prediction, without document verification, does not eliminate the obligation; it merely shifts the compliance risk onto the provider. The ANPD published a technical note in July asking generative AI platforms to demonstrate a legal basis for processing data of adolescents aged 13 to 17, a range that Brazilian law treats under assent and not full consent. ChatGPT for Teens arrives in the country without having publicly responded to this request.
The bottom line for technology directors and boards of institutions that already operate integrations with OpenAI's API: the policy on minors is now a real contractual clause, not an institutional promise. Corporate contracts that host consumer-facing chatbots will need to reflect, at a minimum, the same set of guardrails.