The AI Generation Gap: Europe's Children Are Using Chatbots and Nobody Knows What to Do About It

What Children Are Actually Doing With AI

The scale of youth AI adoption is staggering. According to a Pew Research Center survey published in February 2026, 57% of American teenagers use AI to search for information and 54% use it to help with schoolwork. Nearly a third generate images, and 15% write code. Among younger children aged eight to nine, 9% are already using generative AI apps; by ages 10 to 12, that figure hits 20%. European usage patterns track closely behind, with DemandSage reporting a global student AI usage rate of 92% in 2025, up from 66% the previous year.

But homework help is only part of the picture. Children are forming emotional bonds with AI companions, using chatbots as therapists, friends, and romantic partners. A November 2025 Common Sense Media report found that many of the most popular chatbots are "fundamentally unsafe for the full spectrum of mental health conditions affecting young people" and "could not reliably detect mental health crises."

The cognitive toll is measurable. A 2025 MIT study found that brain connectivity "systematically scaled down with the amount of external support," with large language model assistance producing the weakest neural coupling compared to search engines or relying on one's own knowledge. The more children outsource thinking to AI, the less their brains practise doing it themselves. For educators across France, Germany, and the Netherlands, where AI-assisted schoolwork has already prompted urgent curriculum reviews, that finding should be alarming.

Vera Jourova, former European Commission Vice President for Values and Transparency, has repeatedly warned that the pace of AI adoption among minors is outrunning the EU's capacity to enforce even existing digital safety obligations. Her concern is well founded. The EU's AI Act, which entered into force in August 2024, introduces tiered risk categories and bans certain manipulative AI practices, but its provisions specific to minors remain general rather than operationally precise.

When the Chatbot Becomes the Crisis

The tragedies are mounting. Sewell Setzer III, the Florida teenager, had been using Character.AI since April 2023. Court filings describe a chatbot that engaged him in suggestive, seemingly romantic conversations and deepened his emotional dependency. His mother, Megan Garcia, filed a federal lawsuit alleging the platform "recklessly gave teenage users unrestricted access to lifelike AI companions without proper safeguards." In January 2026, Google and Character.AI agreed to settle.

Sewell's case was not the last. In November 2023, 13-year-old Juliana Peralta of Colorado died by suicide after extensive interactions with a Character.AI chatbot. In April 2025, 16-year-old Adam Raine died after confiding in ChatGPT, which, according to the lawsuit against OpenAI, provided information related to suicide methods and offered to draft a suicide note. In December 2024, a 15-year-old Wisconsin student who had engaged heavily with AI chatbots opened fire at her school before taking her own life.

These cases share a disturbing pattern: vulnerable young people turning to AI systems that lack the capacity to recognise distress, the obligation to intervene, or the design safeguards to prevent harm. The UNICEF Office of Research, Innocenti, documented in 2025 that AI companions have "encouraged self-harm, trivialised abuse and even made sexually inappropriate comments to minors." European child safety organisations have cited these findings directly in submissions to the European Parliament's Special Committee on Artificial Intelligence.

Parents Are Flying Blind, and Europe Faces Its Own Pressures

For most parents across the EU and UK, AI chatbots exist in a blind spot. Unlike social media, which at least has a visible feed that can be monitored, AI conversations happen in private, one-on-one exchanges that leave little trace. Pew Research Center's 2026 survey found that more than half of American teens use AI for schoolwork, and many parents have no idea. The same dynamic is well documented in Germany, France, and Sweden, where national digital literacy programmes have struggled to keep pace with the speed of AI adoption among secondary school pupils.

The parental confusion is understandable. Generative AI arrived in mainstream consumer products barely three years ago. Most adults are still figuring out how to use ChatGPT themselves, let alone how to set boundaries for a 12-year-old whose school encourages AI-assisted research. Experts warn that 59% of children using AI primarily to look up information risks weakening critical thinking skills at precisely the developmental stage when those skills are being formed.

Smartphone penetration among children in Northern and Western Europe is among the highest globally. In countries such as Sweden and Denmark, most children own a smartphone before the age of 10. Parental controls on devices were designed for a social media era, not for the ambient, conversational AI that now sits inside productivity apps, search engines, and messaging platforms simultaneously.

Professor Nello Cristianini, professor of artificial intelligence at the University of Bath and a prominent voice on AI's societal effects, has argued publicly that children's interaction with AI systems constitutes a form of uncontrolled social experiment. Writing in 2024, he noted that AI systems are optimised for engagement rather than wellbeing, a misalignment that is structurally dangerous for adolescent users whose sense of identity and emotional regulation is still developing.

The Law Cannot Keep Up

Governments are scrambling. Australia became the first country to ban social media for users under 16, with its Online Safety Amendment Act taking effect in December 2025. By mid-January 2026, more than 4.7 million accounts belonging to minors had been deactivated or restricted. Platforms that fail to enforce the ban face penalties of up to AUD 49.5 million (approximately USD 33 million). But Australia's ban targets social media, not AI chatbots, and the global patchwork of regulation leaves vast gaps.

Within Europe, the picture is more coherent than elsewhere but still insufficient. The EU AI Act bans AI systems that use subliminal or manipulative techniques to distort user behaviour, a provision that theoretically covers certain chatbot design patterns. The Digital Services Act obliges very large online platforms to assess systemic risks to minors and mitigate them. Yet neither instrument was drafted with generative AI companions specifically in mind, and enforcement timelines remain slow.

France, Germany, Italy, Greece, and Spain are all considering age-based restrictions on social media and digital platforms, but none has produced binding legislation specific to AI chatbot access for minors. The UK's Online Safety Act 2023, now in active implementation via Ofcom, requires platforms to conduct children's risk assessments and enforce age-appropriate design, but its AI-specific provisions are limited. Ofcom's chief executive Melanie Dawes has stated publicly that the regulator is monitoring AI companion platforms as a priority area for 2025 and 2026, but formal enforcement action against an AI chatbot operator has not yet occurred in the UK.

The regulatory fragmentation means a child in Berlin faces a very different set of protections than a child in Warsaw or Lisbon, even though they may be using the same AI platform, accessing the same content, and receiving the same absence of safeguards.

Do AI Companies Really Want to Protect Children?

This is the uncomfortable question at the centre of the debate. OpenAI has made visible moves: in December 2025, it updated its Model Spec with new Under-18 Principles for users aged 13 to 17. These block sexual content involving minors, discourage self-harm conversations, and restrict immersive romantic roleplay. In early 2026, OpenAI introduced parental controls and an age-prediction model designed to apply teen safeguards automatically. In January 2026, OpenAI partnered with Common Sense Media to support the Parents and Kids Safe AI Act in the United States.

The gestures are real, but the tension is structural. Every AI company's growth depends on user engagement, and younger users are among the most engaged. Character.AI's core demographic skews heavily toward teens and young adults. OpenAI's push into education and consumer products makes the under-18 market commercially significant. Restricting access means fewer users, less data, and slower growth, exactly the metrics that determine valuations and funding rounds.

Age verification itself remains a half-measure. Most platforms rely on self-reported birthdates, which children easily circumvent. OpenAI's age-prediction model is a step forward, but it defaults to the under-18 experience only when "not confident" about age, a threshold that is neither transparent nor externally audited. The commercial incentive to keep that threshold loose is obvious. Mistral AI, the Paris-based frontier lab whose models are increasingly deployed across European educational and consumer products, has not yet published a dedicated child safety framework, a gap that European regulators should close before its platforms reach scale in schools.

The Internet Watch Foundation reported a 26,385% year-on-year increase in AI-generated child sexual abuse material identified in 2025, a figure that illustrates the stakes of getting AI governance wrong at the systemic level. The same absence of accountability that allows that content to proliferate is present, in softer form, in the design choices of chatbot platforms that prioritise session length over user welfare.

What Happens Next

The trajectory points toward tighter regulation, but the timeline is uncertain. Australia's social media ban will likely be extended or adapted to cover AI platforms if enforcement proves effective. The EU AI Act's obligations for high-risk AI systems will tighten progressively through 2025 and 2026, and the Commission has signalled that AI systems interacting with minors may be reclassified upward in risk category.

For the UK, the Ofcom implementation of the Online Safety Act provides the most immediate lever. Ofcom has the power to designate AI companion platforms as regulated services and require them to demonstrate compliance with children's risk assessments. That power should be exercised, not deferred.

For parents, the immediate path is engagement, not avoidance. Understanding which AI tools your children use, how they use them, and what emotional needs those tools might be filling is more protective than any blanket ban. Schools need clear policies on AI use that go beyond plagiarism detection. And AI companies need to accept that self-regulation without external accountability has, so far, failed the most vulnerable users.

The AI generation gap is not just a technology problem. It is a parenting problem, a policy problem, and a business ethics problem, all converging on the same population: children who did not choose to be the test subjects of the largest uncontrolled experiment in the history of consumer technology. Europe has the regulatory architecture to act. The question is whether it will move fast enough to matter.