‍

As artificial intelligence (AI) promises so much opportunity, how can we ensure that vulnerable groups, particularly children, are able to realize its benefits while remaining protected from potential harm? Social media offers a powerful lesson in how technologies with insufficient protections can expose children to unacceptable risks and can cause lasting harm. A solution is rooted in policies that protect minors from high-risk AI features alongside an intentional K-12 curriculum that develops minors’ AI literacy.

‍

From our experience with social media, we know that children are disproportionately affected by these risks. Some of these harms include a doubled risk of poor mental health, including depression and anxiety, low self-esteem, and exposure to inappropriate content. Certain AI features lend themselves to causing similar issues for children. As Common Sense Media recently found, “kids who report feeling lonelier tend to use AI more frequently and use AI more often for social and emotional support” suggesting that young people may increasingly rely on AI for socioemotional well-being, even though AI systems are not designed for these purposes. Generative AI features such as human-like companion chatbots, hyper-realistic image manipulation, frictionless emotional validation, and addictive engagement loops are a few of the AI features that pose similar risks of psychological and physical harm to minors.  

‍

From the public more generally, there is healthy skepticism across age groups about the impact of AI on society. A recent Pew survey found that more than half of Americans (57%) rate the societal risks of AI as high, with the most common concern being the potential for AI to weaken human skills and connections. These well-founded concerns show that, unlike with the introduction of the world wide web to the public in 1993, the public is now reluctant to buy into the “wild west” or “move fast and break things” mentality of past technology disruptions. Yes, the public wants guardrails that preserve the benefits of new technologies, but they also want to mitigate potential harms, many of which we have seen from the failure to act on protections for vulnerable groups as they engaged with social media.

‍

Many AI companies recognize these risks and some have implemented guardrails, such as age requirements. OpenAI, for example, launched under-18 (U18) safety policies in November 2025, stating that these protections would apply by default if the user’s age is unknown. Anthropic requires users to be 18 years old to access Claude, likely because it does not offer a version with youth-specific safeguards or parental controls. Google Gemini does not have a minimum age, but it requires parental consent for children under 13. To be effective, these age restrictions must be made hard to circumvent and enforceable. These differing approaches, while potentially positive, reflect a lack of coordination and consistent standards for protecting youth on AI platforms.

‍

With protections in place, how then do we also ensure that kids learn to challenge questionable assertions made by AI, when AI claims to review a vast expanse of human knowledge and synthesize that into “answers?” For those of us from the pre-AI period, we purposefully learned research and critical thinking skills to develop hypotheses, walking through the scientific method or logical reasoning to form and test our conclusions. For the generation being raised in an AI environment, we need to rethink the ways we teach those skills and ensure a setting where those skills can be learned,  internalized, and maintained.  

‍

As a policy matter, state and local boards of education should consider modifications to their curriculums to more explicitly teach these skills at each level of K-12 learning. At the federal level, policymakers should consider placing restrictions on certain features of AI tools for children, so they have the opportunity to develop critical thinking skills. A framework that both limits minors’ access to certain features of AI tools, and places oversight and enforcement authority with a federal agency, is key.  

‍

A majority of state governments are starting to develop frameworks that address the role of AI at the K-12 educational stage. These developments cannot come soon enough, as teachers are left to navigate AI training independently in the absence of clear guidance. Maryland, for example, just adopted a framework to help shape the role of AI in education for its students and schools. It is designed to support local education agencies, schools, and partners in making informed decisions about if and how AI is used in teaching, learning, and school operations, while affirming that technology must serve educational goals and that human judgment, relationships, and student wellbeing remain central. Georgia has adopted a similar framework that recognizes the value of AI while committing to the understanding that AI is a tool to enhance, not replace, human judgment and interaction. These are encouraging developments that point to the fact that states are stepping up to promote ongoing critical thinking.

‍

At the federal level, there are promising signs as well. Just last week, the Senate Commerce Committee took up bipartisan legislation led by Sen. Ed Markey (D-MA) that would put in place transparency and disclosure requirements, as well as restrictions on the use of information collected on children. These are strong first steps, but more will need to be done. An important next step is charging a specific federal regulatory agency with adopting and evolving rules (or enforceable codes of conduct) governing dissemination of AI to minors.

‍

Recently, a group of high school students from across the country passed mock legislation that would limit AI usage to “editing, brainstorming, and studying” and required that students learn how to cite and “demonstrate their grasp of a subject through discussion, handwritten tests or oral defenses.” This approach recognizes AI as a learning tool rather than a substitute for learning. State and federal policymakers should take note and put some rigorous thought to the role AI is to play and adopt appropriate policies to advance those objectives.  

‍

Having lived through the “comprehensive privacy law” effort for the last 20+ years with no result to show except harm to children, we need policymakers to focus on achieving meaningful change, no matter how incremental, to bring about meaningful AI protections. This will require development and implementation of holistic AI policies focused on the three Ss – safety, supervision, and skills. No AI policy platform will be complete without addressing all three.

‍