Finally, some consequences for Big Tech’s careless manipulation of young minds. A surge of trials in the US is setting a precedent for holding social media companies accountable for designing platforms that harm children. But the route here is paved with generations of kids addicted to destructive products. 

Now, a new wave of emerging tech is flooding young people’s lives, and the law isn’t keeping up. Governments need to course-correct by regulating AI use in K-12 education before these systems cause lasting harm to children's learning.

A long road to accountability

Educators and families witnessed firsthand as children's attention was mined by Instagram, Snap, and TikTok, among other platforms. They’ve spoken out on how these apps rip kids from real human connection, fracturing their safety and well-being in the process. Young people themselves have described how social media makes them feel worse about themselves and damages their mental health. In some cases, they wish their preferred platforms were never invented, but feel tethered to them nonetheless.

For years, public consensus was an insufficient match for the financial might of Big Tech, but the law is catching up. A jury in New Mexico found Meta knowingly deceived consumers about product safety, harming young people in the process. A judge in the same case ruled the company was responsible for a public nuisance—in part because of its impact on children’s mental health. The company was ordered to pay $942 million and must meet binding design requirements.

Then, at the end of August, Meta settled with 47 states, the District of Columbia, and three territories as a result of the bellwether trials accusing the company of fueling social media addiction among young users. Without admitting wrongdoing, Meta will pay up to $17 billion over 10 years and make design changes applicable to minors in the US. 

But while social media is under intense scrutiny, the same industry is, again, shoving untested technology into children’s palms despite the risks.

Cognitive harms of generative AI

Social media and AI chatbots share a design imperative to maximize user engagement. Social media addiction manifests through features like infinite scroll, public like counts, and algorithmically tailored content feeds. AI chatbots may draw users into extended interactions with sycophantic and anthropomorphic characteristics, unpredictable responses, and answers which discourage users from ending conversations. Young people are susceptible to dependence on systems built to capture their attention, increasing exposure to harm.

But unlike social media, AI chatbots are explicitly marketed as an educational tool for children, despite threatening their cognitive development.

Brookings analyzed research on how AI impacts learning and, though the evidence base is small, concluded that risks such as cognitive offloading currently outweigh potential benefits. Generative AI may reduce knowledge retention: in China, among 26,811 students aged 12-18, those who used AI for schoolwork performed better on homework, but worse on tests. Other researchers have warned that AI use may diminish critical thinking and creativity.

A recent review cautions about the claims that ed tech leads to positive learning outcomes. Benefits are highly dependent on the careful alignment between children’s development, a product’s design, and the pedagogy surrounding its use. This may explain why a randomized control trial found that high school students who used a standard GPT for math practice scored lower on their exams. Those who practiced with a version designed with pedagogical guardrails performed similarly to students who used no AI at all.

Regardless of what studies are telling us, how do teachers and students feel? In a survey of 4,423 educators in British Columbia, 86 percent worried about students’ over-reliance on AI, with 83 percent concerned it could negatively impact critical thinking. Over three-quarters felt AI may weaken educators’ ability to gauge what children understand, reflecting research on how AI systems fracture teacher-student relationships.

Common Sense Media surveyed 1,017 American teens (aged 13-17) and found 70 percent use AI tools for schoolwork, most frequently AI chatbots. Of this group, 63 percent reported getting answers from AI, outsourcing their cognitive effort instead of working through problems themselves. Uptake among students is high, but so is skepticism. Children are open to AI as an educational tool but want decision-makers to address safety risks, over-reliance, and negative learning outcomes.

Deployment at all costs

Despite signals of cognitive harm, companies like Google, Microsoft, and OpenAI are marketing AI chatbots to students and schools, promoting hypothetical benefits for learning. OpenAI claims the Study Mode feature of ChatGPT for Teens is grounded in learning science principles. But applying principles to an AI chatbot is a far cry from rigorously testing a product before scaling it to children globally. The company points to its own evaluation, but this was conducted with 300 college students, not the 13-17 year-olds they’re pitching to, and gains were limited anyway.

Google makes similar claims about the scientific basis of its school-based chatbot, but its 2025 evaluation of Gemini as a learning tool was done with educators, not students. The learning science principles the company references are sound, but did not stem from research on kids interacting with algorithms; the pedagogical principles they scored their model against, like active learning and metacognition, have long been validated through research on human instruction. Is it really possible for a chatbot to simulate the challenge and friction young people need to learn? The answer may be yes, but we need more research to know for certain. Google’s own paper concedes the likely limitations of AI in satisfying core relational dimensions of learning. Until this open question is answered, tech companies should not be able to market chatbots to schools and young students, profiting massively based on false promises. Regulation can stem the tide.

Enforceable safeguards for AI in K-12

The Independent International Scientific Panel on AI warned that AI models are developing faster than governance. This is certainly the case in K-12 public education, at great risk to children’s learning. Proposed online safety legislation in Canada and the US takes important steps toward ensuring AI systems used by minors are more safely designed, but governments must ensure laws cover platforms intended for learning.

More evidence is needed to fully understand how to regulate AI used in K-12 education to mitigate cognitive harms. Independent studies on how AI shapes children’s learning are urgently needed, and must go beyond exam scores to competencies students need to thrive, like critical thinking, creativity, and collaboration. Crucially, affected groups must be engaged in determining the path forward. That means students and their families as well as educators and their unions.

Waiting for research to catch up is not an option. Some steps governments should take in the meantime include:

• Designating AI systems used in K-12 education as high-risk and scoping them into online safety legislation. In particular, AI tutors and chatbots, widely marketed to schools and students, must be regulated given their significant potential to impact the right to education. Products should be regulated based on how they are marketed and used, regardless of whether they are formally procured by schools.
• Defining design standards that mitigate risks to cognitive development. Companies should be required to meet safety-by-design and pedagogical-by-design standards, defined with school communities and informed by evidence on what supports learning (e.g. fit for age, developmentally appropriate design, suitability for pedagogy). Features that work against learning and healthy development—like frictionless answers, engagement optimization, and anthropomorphic personas—should be prohibited.
• Generating independent evidence. AI used in K-12 education should be independently evaluated to validate claims about learning outcomes. Testing must be done on the group the product is targeted to, with results published to enable informed procurement and engagement decisions.
• Enforcing accountability mechanisms throughout the life cycle of an AI product used by children for learning. Child rights impact assessments should be mandatory before deployment and after any design change. Companies should be subject to ongoing safety audits, the results of which should be public. And there should be accessible routes for young users to complain and seek redress for harms.

Where applicable, grounding these measures in ed tech procurement standards would ensure all school districts, regardless of resources, are equipped to hold tech companies accountable.

If governments ignore the risks, the sequence that has just played out for social media platforms will eventually repeat for companies profiting from student-facing AI products that cause cognitive harms. The pattern is consistent: an enticing digital platform is widely adopted, evidence of impacts to young people emerge, tech companies deflect responsibility for harms, putting the burden on parents to keep kids safe online, and only after years of damage do courts catch up.

Now, educators are being held responsible for safeguarding students from the complex cognitive risks associated with AI platforms, instead of the companies designing and profiting off the products. In the case of social media, millions of affected minors and mass public outcry were eventually enough to compel legal consequences. Let’s not wait for a generation to lose out on their right to education before we demand action. Instead of waiting for AI companies to be sued into submission, governments must do their job and regulate.