Summary: For the first time, a generation is coming of age alongside AI. Product teams should keep these close: they convert “safety” from an abstract promise into a set of testable design criteria. OpenAI’s July policy essay leads with the headline figure: of the teens on ChatGPT, nearly 9 in 10 turn to it every week to learn, look things up, build skills, or get more done.
For the first time, a generation is coming of age alongside AI. OpenAI’s July policy essay leads with the headline figure: of the teens on ChatGPT, nearly 9 in 10 turn to it every week to learn, look things up, build skills, or get more done. The company’s stance leaves little room for misreading — holding teens back from AI until adulthood would be the equivalent of telling an earlier generation to stay off the internet and search engines until 18, leaving them underprepared for the defining technology of their era. Access, however, has to arrive bundled with protections built for this specific age group. As a human anchor, the essay points to Oriana McKenzie, who told an OpenAI Academy audience that ChatGPT helped her study smarter and keep her workload manageable — enough to reach the top of her high school class while still making room for the sports and programs she loves.
Four Governance Principles: Treating Teens as Teens
The past year at OpenAI has brought hardened default protections, the rollout of age prediction, expanded Parental Controls, family-facing resources, and learning features that reward understanding over quick answers. Underneath sits a set of four stated principles, quoted nearly verbatim: “We put teen safety first even when it may conflict with other goals. We encourage real-world support in times of need. We treat teens as teens. We are transparent by setting clear expectations.” Each one shows up in product behavior — safety overrides engagement metrics, crisis moments route toward real people rather than in-app placation, defaults assume a still-developing user, and settings describe plainly what they do. Product teams should keep these close: they convert “safety” from an abstract promise into a set of testable design criteria.
Building for Learning, Not Just Answers
Where teens benefit from AI most clearly is learning. Study Mode — built together with teachers, learning scientists, and pedagogy experts — moves students through problems using guiding questions, structured explanations, and moments of reflection, rather than handing over the finished answer. OpenAI points to early evaluations with promising learning gains, and folds those findings into wider ongoing research. For parents, the friction is now lower: Study Mode can be switched on straight from Parental Controls for a linked teen account, and once enabled it becomes the default for every new chat.
The supporting cast includes education-oriented starter prompts (split a topic into steps, convert notes into a study guide, generate flashcards and practice questions, check evidence and clarity) plus a growing library of interactive learning experiences — 18 million weekly users now touch the interactive math and science content, which spans more than 300 topics from integrals and mitosis to moon phases, alongside a pronunciation experience covering more than 61 languages. The through-line never changes: research keeps confirming that active engagement beats passive consumption for learning, so the product keeps leaning into interaction.
Built-in Safeguards: Age-Appropriate Experiences and Healthy Habits
Whenever OpenAI’s systems estimate that a user is under 18, a more age-appropriate experience kicks in automatically: learning, creating, and exploring remain fully available, while guardrails tighten around graphic violence, self-harm, risky viral challenges, unhealthy body-image content, and dangerous, romantic, or sexual roleplay. The stated North Star is blunt — ChatGPT is meant to stay a tool for learning and creativity, not a stand-in for real-world relationships.
Healthy usage patterns got product treatment too: teens who spend long continuous stretches in ChatGPT see break reminders arrive more often, nudging them to pause and step away from the screen. Freedom to use AI and balanced use are treated as compatible goals, not opposing ones.
Parental Controls: From Tools to High-Risk Notifications
On the parent side, the toolkit covers quiet hours, disabling voice mode, managing image-generation permissions, and notifications for high-risk situations — for instance, indications of potential self-harm. That notification surface has now widened: when a linked teen account gets deactivated for violating usage policies around violent threats or online violence, parents are told as well. The intent is to make sure families know when something serious happened, while still respecting the teen’s privacy and steering the conversation and support offline. To keep these interventions thoughtful and effective, OpenAI consulted outside experts including Moonshot, an organization focused on preventing online violence.
Outside Experts and Industry Standards
No single company can carry teen online safety on its own. OpenAI’s collaborators span teens, parents, educators, child-safety and mental-health specialists, researchers, governments, and civil society — and this year it joined the Family Online Safety Institute (FOSI) to push shared standards built around empowered parents and responsible industry practice.
What Product Teams Should Take Away
Going forward, OpenAI says it will keep tightening age-appropriate protections, equip parents with more tools and control, strengthen safeguards against serious harms, deepen research on healthy AI use, and ship more experiences that help teens learn actively and use AI with confidence — working alongside teens, parents, educators, experts, and communities worldwide.
For builders, the payoff of this essay isn’t the feature inventory — it’s the ordering of priorities: access and protection are welded into a single commitment; age prediction routes users into age-appropriate defaults rather than optional add-ons; learning features tilt toward guided understanding instead of direct answers; and parental tooling stretches from control (hours, feature toggles) to awareness (high-risk notifications), engineered to honor privacy while prompting offline support. Any team building for teens can lift these four trade-offs directly as design criteria.