OpenAI Launches ChatGPT for Teens
OpenAI has launched ChatGPT for Teens, aimed at users aged 13 to 17, integrating stricter safety restrictions, learning tools, and parental controls into a standalone experience.
This mode restricts high-risk conversations involving self-harm, violence, eating disorders, explicit content, and graphic images, and reinforces rules that prevent the model from suggesting it has personal feelings. In limited cases where serious safety risks are identified, the system can send necessary safety notifications to associated parents' accounts.
OpenAI requires registrants to declare their age and deploys an age prediction model to automatically determine if an account likely belongs to a user under 18. The model integrates over 2,000 behavioral and account signals, including account duration, common active hours, long-term usage patterns, and declared age; if an adult is mistakenly categorized under the teen mode, they can switch back after verifying their age through a Persona selfie.
In terms of educational design, when the system detects that a teen is completing homework, it prioritizes follow-up questions, guided reasoning, and step-by-step understanding instead of providing direct answers. OpenAI is also collaborating with CodeAI to promote AI learning support for students in some schools in the U.S., attempting to position the chatbot as a learning aid rather than just an answer generator.
Once parents are linked to teen accounts, they can set quiet hours, disable voice mode, memory functions, or image generation, and choose not to allow teens' conversations to be used for training the ChatGPT and Sora models. Parents cannot view specific chat records of teens by default and will only receive limited notifications when potential serious safety risks are identified through system and manual reviews.
Market mechanism: The buyers are families, schools, and educational institutions seeking generative AI with content protection and learning assistance for teens; the sellers are OpenAI and competing platforms offering AI products for minors. The event is driven by controversies over teen safety, regulatory scrutiny, and schools' demand for homework assistance tools, with the product using age identification to route underage users into restricted model pathways; OpenAI gains user entry points in family and educational scenarios, while content review, identity verification, and educational technology service providers benefit, relying on general chat products without age thresholds or content rating experiences.
Source: Public information
ABAB AI Insight
Before launching the teen product, OpenAI has continuously strengthened protections for minors: its Model Spec has incorporated principles for users under 18, with plans to further integrate local crisis intervention resources, provide one-click hotlines, increase rest reminders, and collaborate with over 260 mental health experts to improve the model's recognition of distress and self-harm signals by 2026. This release does not merely add a "child mode" button but integrates age inference, content filtering, crisis escalation, and parental account connections into the default product architecture, responding to ongoing controversies regarding generative AI's potential to cause dependency, provide erroneous suggestions, and expose minors to inappropriate content.
From a capital perspective, the age prediction model transforms "user age" from a registration field into a commercial infrastructure that influences model behavior, available features, training data permissions, and risk notification pathways. Parents can disable memory, voice, image generation, and model training usage, meaning that the data availability of teen users is split into service permissions and training permissions; OpenAI exchanges lower data collection flexibility for family customer trust, school deployment possibilities, and reduced compliance risks. Persona's selfie age verification shifts the friction of misjudging young users to third-party identity verification capabilities.
Analogies can be drawn with YouTube Kids, Instagram Teen Accounts, and Apple's Screen Time: platforms first reduce risks for minors through account tiering and parental configurations, then place recommendations, direct messages, content production, or usage duration within controllable boundaries. OpenAI's distinction lies in that ChatGPT not only distributes content but also generates answers, suggestions, and emotional responses in real-time, thus risks arise not only from existing content libraries but also from the model's immediate behavior in specific conversations. The industry has shifted from "adding age prompts before generative AI" to a control phase of "dynamically routing models and permissions by age."
This represents a regulatory change driven product architecture reconstruction. For minors, platforms cannot merely prove that content is safe in principle; they must also be able to identify users, limit functionalities, maintain necessary safety handling links, and allow guardians configurable intervention rights. The accuracy of age identification, the cost of misjudgment appeals, content boundaries, and crisis escalation rules will become the infrastructure for AI companies to gain trust from schools, families, and regulators; those who can embed safety requirements into the default product pathways will find it easier to gain access to the next generation of high-frequency users.
ABAB News · Cognitive Laws
The entry point for minors is not a traffic entry but a responsibility entry.
Age is not a label; age determines product boundaries.
AI must first learn restraint before entering education.