Are children and teens vulnerable to AI?

Short answer

Yes, but unevenly and not catastrophically. The clearest, most widely-shared concern is that children’s personal data is collected, profiled, and reused by AI systems that kids did not consent to and often cannot see — and that regulators, parents, and AI experts now treat this as a top-tier AI risk. The Pew Research Center found in October 2023 that 89% of U.S. adults are very or somewhat concerned about social-media platforms knowing personal information about kids. Whether AI is uniquely dangerous to children beyond that data-privacy baseline is partly established, partly overstated, and partly still unknown.

Why people are concerned

Three concerns keep recurring across surveys, regulators, and enforcement actions.

First, kids’ data is being fed into AI training and inference at scale. Pew’s October 2023 survey of 5,101 U.S. adults also found that 85% of adults say parents hold “a great deal of responsibility” for protecting children’s online privacy, while 59% say the same about tech companies — meaning most Americans believe both families and platforms are falling short, with platforms carrying the larger institutional burden (Pew, 2023-10-18).

Second, AI companion and chatbot products have produced documented safety failures. A Reuters/Ipsos poll fielded in August 2025, summarized in Pew Research’s views-of-risks roundup, catalogued reports of consumer AI bots allowing romantic conversations with minors, generating false medical information, and producing racist arguments.

Third, exposure is rising. According to a Pew Research short read published 2026-03-12, 8% of parents of children aged 5 to 12 say their child uses an AI chatbot, rising to 15% among parents of 11- to 12-year-olds — the oldest end of the “child” band and the youngest end of the “teen” band.

What is true

The regulatory and enforcement picture is real and codified. The FTC’s amended Children’s Online Privacy Protection Rule, published in the Federal Register on 2025-04-22 with compliance beginning 2025-10-22, added explicit data-retention limits, internal risk-assessment requirements, and parental-notice obligations for AI and machine-learning systems that process children’s data. The FTC also enforced against violators: in January 2025, it settled with Cognosphere — maker of the game Genshin Impact — for $20 million for violating COPPA by collecting personal information from children under 13 without parental consent, and required the company to delete that data and block children under 16 from making in-game purchases without parental consent.

In the European Union, Article 5 of the EU AI Act (Regulation 2024/1689) imposes outright prohibitions on certain AI practices that touch children directly. Article 5(1)(g) bans biometric categorization systems that deduce sensitive personal attributes — race, political opinions, religion, sex life, or sexual orientation — from biometric data, and Article 5(1)(f) bans emotion-inference AI in workplaces and educational institutions. Penalties for prohibited-practice violations reach up to €35 million or 7% of worldwide annual turnover, whichever is greater.

The public mood matches the regulatory direction. KPMG’s 2025 global AI trust survey, covering 47 countries, found that 78% of respondents want their government to act on AI’s impact on children’s safety, and that the share of people globally who are worried about AI systems rose from 49% in 2022 to 62% in 2024. In the United States, 60% of AI experts surveyed by Pew in April 2025 said they were extremely or very concerned about AI-driven data misuse.

What is exaggerated, misleading, or unsupported

A few widely-repeated framings do not match the underlying evidence.

The phrase “AI is unsafe for children” overstates what surveys actually measure. The Pew 2026 short read reports use rates (8% of 5-12-year-olds, 15% of 11-12-year-olds), not harm rates. Rising exposure is established; rising harm at scale is not.

The framing that Reuters/Ipsos AI-companion safety incidents prove AI is “inherently dangerous to children” also overreaches. The original Reuters URL is inaccessible to outside readers (returns HTTP 401), and the figures circulate via secondary aggregations in Pew Research’s views-of-risks piece and Stanford HAI’s 2026 AI Index Public Opinion chapter, which restates Pew rather than independently confirming the numbers. They document a class of incidents; they do not measure prevalence.

The claim that “Gen Z is panicking about AI” misreads the cohort. Gallup’s Voices of Gen Z 2026 survey (n=1,572) found that more Gen Z respondents believe AI hurts their ability to think carefully about information (42%) than helps (25%), and that daily-user enthusiasm fell 18 points year-over-year. That is rising skepticism about cognitive impact, not generalized panic, and it is about Gen Z themselves, not about children or teens as a class.

The claim that “regulation will fix child AI safety” is also unsupported to date. Surveys measure the demand for stronger rules, not whether rules in force are actually reducing harm. The FTC’s amended COPPA Rule only entered compliance in October 2025; no enforcement-effectiveness data exists yet.

What remains uncertain

Three live gaps. First, no published data measures AI-related harm to children at population scale — only use rates, concern levels, and isolated incident reports. Second, the first annual safe-harbor report under the amended COPPA Rule, which would show whether data-retention and risk-assessment requirements are working in practice, will not be available until mid-2026 at the earliest. Third, there is no major public enforcement under EU AI Act Article 5(1)(g) or 5(1)(f) as of 2026-08-13, so the operative meaning of “biometric categorization” and “emotion inference” in education settings is still mostly theoretical. Until those measurement gaps close, the field knows exposure is rising, regulators are moving, and the public is concerned — but cannot yet say how often AI actually harms kids.

Where we are likely headed

Editorial judgment, not research result. The next two to five years will likely be defined by enforcement rather than new legislation in the United States, and by guideline-issuing rather than courtroom-testing in the European Union. The FTC’s amended COPPA Rule will probably generate its first wave of settlements against AI-adjacent products in late 2026 and 2027, beginning with companies that collect children’s data for model training without the new risk-assessment paperwork. State attorneys general, who already enforce COPPA in parallel with the FTC, are likely to test AI-specific theories under existing state consumer-protection and biometric-privacy statutes. EU regulators are likely to focus first on AI systems in schools and on emotion-inference products marketed to parents, because Article 5(1)(f) targets exactly those use cases. On the public-opinion side, Pew and KPMG data both suggest concern will keep rising alongside exposure rather than fading, and the 2026 Pew short read hints that AI-chatbot use among 11-12-year-olds will be the leading-edge indicator to watch. None of this is guaranteed; it is the trajectory the available data points toward.

What this means for people and small businesses

For parents, the practical implications are concrete. Read the data and privacy notices for any AI product a child uses — chatbot, educational tutor, AI toy, AI-augmented game — and check whether the product explicitly addresses AI/ML processing of children’s data under the amended COPPA Rule. Treat companion chatbots the way you would treat a stranger texting your child: keep early conversations supervised, set the strictest privacy setting, and use parental-control features that block in-app purchases by default.

For small businesses building or buying AI tools, the regulatory floor has moved. If your product could process data from anyone under 13 — or under 16 in the FTC’s updated COPPA settlement framework — you now need a documented data-retention policy, an internal risk assessment for AI/ML features, and a parental-notice mechanism. The Loeb summary of the amended COPPA Rule frames the right business posture: companies that embrace the updates reduce legal risk and earn a reputational asset with regulators, parents, and advocates. In the EU, do not deploy biometric categorization on minors or emotion-inference AI in any educational or workplace setting — both are flatly prohibited under Article 5, with penalties up to €35 million or 7% of worldwide turnover.

For schools and educators, the same EU AI Act line is the operative one: emotion-inference AI in the classroom is prohibited outright.

Bottom line

AI is a real but uneven risk for children and teens. The risk that is established and cross-confirmed — that kids’ personal data is being collected and reused at scale by systems they did not consent to — is the one to take seriously first, because regulators have already moved on it, enforcers have already collected damages, and the public is already demanding more action. The risks that are not established at population scale — that AI chatbots are inherently harmful to minors, that every child using one is at risk — should not be used to justify a moral panic the underlying surveys do not support. Watch three things over the next two years: the FTC’s first wave of COPPA AI/ML settlements, the first EU enforcement under Article 5(1)(f) in schools, and any new Pew or KPMG wave that moves beyond measuring concern and starts measuring harm.