OpenAI is using the G7 stage to push for something we have not really seen before in tech policy: shared, global rules for how artificial intelligence should treat kids and teens, plus a new international institute whose sole job is to keep youth safety from becoming an afterthought in the AI arms race. It is a call that sits somewhere between product roadmap, public policy memo, and open invitation to regulators to hold AI companies to a higher bar.

At first glance, this might sound like yet another corporate manifesto in a year crowded with AI ethics statements. But if you look closely, OpenAI is trying to do two things at once: lock in a baseline of rights and protections for under-18s, and convince governments that the answer is not just more bans and age gates, but a shared framework that treats AI as both risk and opportunity. That nuance matters, especially in a world where American parents, Indian teens, and European regulators are all having very different conversations about the same tools.

The timing is deliberate. OpenAI’s new youth push, laid out in its “Advancing youth safety and opportunity through global leadership” statement, arrives just as world leaders gather at the G7 in France, where AI safety will be one of the headline topics. The company is asking those governments to back the creation of an international youth AI safety institute – a standing body that would sit above any single product cycle or election and keep pressure on both companies and regulators to update standards as models evolve.

If you are picturing a new flavor of AI Safety Institute, you are not far off. OpenAI explicitly frames this as an institute that would coordinate evidence, share best practices, and make sure youth protections do not depend on which country a child happens to live in. In other words, a kind of Geneva for youth AI standards, but with a remit that covers everything from product design to data privacy to educational access.

At the core of OpenAI’s pitch is a very simple idea: companies should know when they are dealing with a minor and act accordingly. That sounds obvious, but it is not how most online services were built. OpenAI is pushing for what it calls “effective, privacy-preserving age estimation” so providers can distinguish between adults and young users, then default to stricter protections when they are not sure. Practically, that means no more shrugging and saying “we can’t tell if you are 13 or 33” – and no more pretending that the same general-purpose model is equally appropriate for everyone.

This push did not come out of nowhere. Over the past year, OpenAI has rolled out a Teen Safety Blueprint, including an India-specific version that acknowledges realities like shared devices and uneven access, plus a broader child safety blueprint that covers younger kids. Those efforts introduced ideas such as under-18 specific models, blackout hours for late-night usage, alerts for self-harm risks, and privacy-friendly age estimation – the building blocks of what is now being scaled up into a global ask.

In its new youth safety principles, OpenAI draws a pretty sharp line between what companies should be doing and what has historically been left to parents. The message is clear: the burden should not sit solely on families to manage AI risks. Instead, companies should be compelled to run annual youth risk assessments, looking not just at harms but also at how their tools benefit young people, and then adjust safeguards before those harms show up in the real world. Think of it as a safety MOT test for AI products, repeated every year, with a special focus on kids and teens.

Parental controls are part of the picture, but they are framed as tools within a larger system, not a magic fix. OpenAI argues that parents and guardians should have “accessible, easy-to-use controls” to manage settings like memory, data use, and time limits – essentially the levers they wish social media had offered a decade ago. The key is usability; controls that live three menus deep or require a computer science degree do not count. Companies, the argument goes, need to design these tools as first-class features, not afterthoughts.

Transparency is another recurring theme. OpenAI says families “deserve clear, understandable information” about how companies are protecting young users and what the tradeoffs look like. That means plain-language safety policies, not legalese-heavy privacy notices, and honest explanations of both benefits and risks – for instance, how a homework assistant might help with learning, but also tempt kids to outsource thinking if used badly. In policy terms, it is a call for something closer to nutrition labels for AI experiences.

The proposed standards also deal directly with content and conduct risks, especially around grooming, exploitation, self-harm, and other sensitive areas. OpenAI wants systems to be designed so they refuse to generate developmentally inappropriate material, while also having clear protocols for what happens when high-risk interactions are detected. This aligns with broader moves by child-safety groups that have been pressing tech companies to treat grooming and abuse patterns as safety-critical signals, not just content moderation issues.

One of the more interesting parts of OpenAI’s framework is what it says AI should be used for with young people. The company emphasizes that AI tools “should be designed to support learning, development, and real-world relationships – not replace them.” That might sound like marketing copy, but baked into it is a line that many parents and educators care about: AI tutors and creative tools are welcome, AI best friends or ersatz therapists are not. Companies, OpenAI argues, should draw clear boundaries in domains where healthy development depends on human judgment and professional support.

Data privacy is non-negotiable in this vision. OpenAI calls for a ban on privacy-invasive, targeted advertising to young people, and on selling minors’ personal information. In effect, that brings AI more in line with emerging norms around child-directed services, especially in Europe and parts of the US, where regulators have increasingly treated youth data as a special category. For AI companies accustomed to monetizing via data, this is a significant pivot: if they accept this standard, business models for youth-focused experiences will have to lean more heavily on subscriptions, licenses, or public funding rather than behavioral ads.

Of course, OpenAI is not operating in a vacuum. UNICEF has been publishing guidance on AI and children, calling for child-centered AI that protects data and privacy, ensures non-discrimination, and builds in transparency and accountability. UNESCO has its own AI competency framework for students, aimed at helping countries redesign education systems around responsible AI use. And a joint statement backed by dozens of organizations in 2025 pushed for a child rights-based approach to AI, framing kids’ rights as a baseline constraint, not an optional add-on. OpenAI’s move sits within this ecosystem, but gives it a concrete corporate implementation plan.

The difference is that OpenAI has real leverage over the design of the models millions of young people are already using, directly or via integrations into apps and classrooms. When it talks about under-18 specific behavior, it is not just a hypothetical; the company has already started rolling out under-18 model specs, age prediction models, and parental controls in regions like EMEA. In India, for example, OpenAI’s teen blueprint explicitly addresses shared devices and network realities, making features like parental controls and blackout hours work in crowded, multi-user households.

The proposed international institute is supposed to knit all of this together. In OpenAI’s telling, this body would convene governments, researchers, civil society, and industry, acting as a hub for evidence on what works, and a vehicle for raising standards over time. Picture a place where researchers bring in data on how teens actually use AI, companies share red-team findings on novel risks, and regulators test out draft rules before they become law. Crucially, the institute would ensure that protections are interoperable – so that an audit in one region counts in another, and youth-focused safety features do not fragment into incompatible national variants.

There is also a politics here. By making this push at the G7 and positioning itself as a partner to governments, OpenAI is signaling that it wants to be inside the regulatory tent, not a target standing outside. Its leaders have been talking publicly about working with emerging AI safety institutes and designing regulation that “enables innovation while protecting users,” especially children. In practice, that could mean global standards that look more like technical specifications and shared audits than one-size-fits-all bans – a regulatory model closer to aviation or pharmaceuticals than to the whack-a-mole of social media moderation.

Whether this lands the way OpenAI hopes will depend on who actually shows up. If the institute becomes a club of rich-country regulators and US-based companies, it will struggle to reflect the realities of teens in Lagos, São Paulo, or rural India. Some of OpenAI’s own work, such as localization efforts in India and multilingual model support, suggests it understands the importance of cultural and linguistic diversity. But turning that into governance – giving Global South voices real influence over standards – is a harder problem than tweaking model specs.

Then there is the trust gap. Even the most detailed blueprint does not erase the fact that many parents and educators feel blindsided by the speed at which tools like ChatGPT have entered classrooms and bedrooms. For them, OpenAI’s call for global standards may read as overdue damage control rather than proactive leadership. That is why the emphasis on independent audits and strong accountability mechanisms matters: OpenAI is not just asking to be trusted; it is asking for a system that verifies whether companies deliver on their promises.

If regulators and other companies sign on, the result could be a sort of minimum global floor for youth AI safety. Under that floor, companies would need to (1) identify minors and default to age-appropriate safeguards, (2) run regular youth-focused risk assessments, (3) offer practical parental tools and transparent safety information, (4) design systems to support learning and wellbeing, and (5) protect minors’ data from targeted advertising and resale. None of that guarantees safe AI, but it reshapes what “responsible” looks like in product roadmaps and policy debates.