OpenAI’s new ChatGPT for Teens is designed to give users under 18 a more restricted version of its chatbot, but child-safety researchers say the company still needs to prove that the technical systems behind those protections work reliably.
Engadget reports that the teen experience automatically becomes the default when OpenAI predicts a user is under 18 or the user identifies themselves as a minor, while youth-safety experts interviewed by the publication called for extensive independent testing and greater transparency around the safeguards.
The New York Times reported that ChatGPT for Teens limits high-risk conversations involving self-harm, violence, eating disorders, and explicit sexual or graphic material, while adding study-oriented guardrails intended to guide teenagers through schoolwork rather than immediately supplying answers.
For builders, the rollout turns child safety into an engineering problem spanning identity, model behavior, interface design and moderation operations.
Age Prediction Becomes Core AI Infrastructure
Determining who should receive the restricted experience is one of OpenAI’s biggest technical challenges.
OpenAI uses more than 2,000 signals to estimate whether someone is younger than 18, including factors such as typical login times and how long an account has existed; the company described those signals as “minimally privacy invasive.”
But Engadget reports that Robbie Torney, head of AI and digital assessments at Common Sense Media, said OpenAI has not published false-positive and false-negative rates demonstrating how accurately the system identifies teenagers.
That is important for AI product teams: safeguards only work if the system successfully routes the intended users into them.
OpenAI Is Deliberately Making the Chatbot Less Human
OpenAI is also changing the interaction model itself.
BBC reported that new teen accounts can disable human-like voice responses to reduce the chatbot’s ability to appear like a person, while regular prompts remind younger users that ChatGPT is AI and that “it can wait.”
The New York Times also reported that the teen mode strengthens rules preventing ChatGPT from suggesting that it has personal feelings, alongside restrictions on sensitive conversations.
That signals a wider product-design shift for conversational AI: anthropomorphism itself is becoming a safety variable.
Human Review Creates a Scaling Problem
Some protections depend not only on classifiers but on people.
OpenAI youth and families chief Lauren Jonas said flagged material is reviewed by full-time employees before a linked parent receives a notification, with the company aiming to alert parents within one hour.
That commitment has drawn skepticism. Earlier Common Sense Media testing found some parental warnings took 24 to more than 48 hours, while some expected alerts never arrived.
Safety-by-Design Is Becoming an AI Product Requirement
Fairplay executive director Josh Golin argued that stronger defaults are more effective than placing responsibility on parents to discover and configure safety controls themselves.
For AI builders, that may be the larger lesson. ChatGPT for Teens is testing whether safety can be designed into model behavior, age estimation, product interfaces and moderation infrastructure rather than bolted on afterward.
If OpenAI can demonstrate those systems work reliably at scale, the architecture could influence how other AI companies build for minors. If it cannot, the industry may find that launching a separate “teen mode” is much easier than proving the machinery behind it is actually safe.
