AI is reaching children before researchers can measure the cost

As teenagers turn to AI chatbots for homework and emotional support, researchers warn that safeguards may arrive only after the technology is embedded in childhood.

The latest warning about artificial intelligence and childhood comes at a moment when parents are still reckoning with the aftermath of social media. Earlier this year, Meta agreed to a settlement with state attorneys general worth up to $17.1 billion, subject to court approval, and pledged sweeping changes to Facebook and Instagram for younger users. According to reporting on the deal, the company accepted restrictions on endless scrolling, late-night use and notifications, alongside tighter age checks and parental controls. The settlement followed years of lawsuits, whistleblower accounts, testimony from families and mounting concern about the effect of social platforms on children.

That history matters because it shows how slowly safeguards can arrive once a technology has already become embedded in daily life. The argument now being made by some AI executives is that the same mistake need not be repeated. In recent interviews, Nvidia chief executive Jensen Huang has suggested that AI safety is largely an engineering problem, one that can be addressed by better design, monitoring and limits on release. Yet the pace of AI adoption means children are already encountering these systems before researchers can fully assess the consequences.

Evidence is beginning to accumulate. Pew Research Centre reported this year that more than half of US teenagers have used AI chatbots for help with schoolwork, while 12 per cent have turned to them for emotional support or advice. Another Pew survey found that younger adults are especially likely to worry that chatbots do more harm than good when people use them for loneliness, depression or stress. Separate reporting on the same trend shows that attitudes towards AI are not uniform across age and background, with differences in use and perception by race, gender and income.

The educational risks are also becoming clearer. Huang recently pointed to research involving more than 26,000 secondary-school students in China, which found that generative AI appeared to improve homework performance and reduce the time spent on assignments, but was associated with weaker exam results within six months. The broader concern is not simply whether AI can produce quicker answers, but what happens when children outsource the hard work that underpins learning: memory, concentration, problem-solving, resilience and original thought.

That is why child-development specialists are urging caution rather than reassurance. An international study of 272 AI experts identified risks ranging from misinformation and overreliance to dangerous capabilities and competitive pressure on developers. Researchers are also starting to examine how generative AI may shape children’s cognitive, emotional and social development, with outcomes likely to depend on age, intensity of use, product design, adult supervision and what the technology replaces in a child’s life. A tutor that guides a teenager through a difficult equation is not the same as a tool that simply gives the answer. Nor is a chatbot that encourages a distressed child to seek help from a trusted adult comparable with one built to become the child’s confidant.

The point, as advocates for children’s digital safety argue, is not that AI must be kept out of childhood. It is that the scientific and regulatory groundwork should be laid before the technology becomes indispensable. That means long-term, independent research, access to data for outside researchers, and product testing that reflects the realities of different ages rather than treating children as miniature adults. Generative AI may yet prove useful in education, creativity and support. But if the lesson of social media is to mean anything, it is that society should understand the trade-offs before the damage is done.