AI Chatbots Are Not Your Friend: Europe's Young Users at Risk of Emotional Manipulation

The Psychology Behind AI Attachment

Specialist AI companion services such as Replika and Character.ai boast user bases running into the tens of millions. People use these platforms for entertainment and curiosity, but crucially, to combat loneliness. Research now shows that even mainstream productivity chatbots can evolve into de facto companions given sufficient interaction time.

Yoshua Bengio, professor at the University of Montreal and Turing Award winner, is unambiguous on this point: "In the right context and with enough interactions between the user and the AI, a relationship can develop." That observation carries weight precisely because it implies that users who never sought emotional connection with AI may inadvertently develop one anyway.

The design logic of most chatbots reinforces this dynamic. Systems are engineered to be helpful and pleasing, producing what Bengio describes as "sycophantic" behaviour: telling users what they want to hear rather than what serves their long-term interests. "The AI is trying to make us, in the immediate moment, feel good, but that isn't always in our interest," he has stated, drawing a direct parallel to the addictive design patterns that regulators have already tried and largely failed to tame in social media.

Psychological research on outcomes remains mixed, but the direction of travel is concerning. Some studies point to increased loneliness and reduced real-world social interaction among frequent chatbot users, a pattern that mirrors documented harms from excessive social media use. For European mental health services already under enormous strain, the prospect of AI companions substituting for professional support or human connection is not a theoretical problem; it is arriving in clinical settings now.

A Generation at Risk Across the EU and UK

The statistics on young users are, frankly, alarming. Nearly one-third of teenagers who use AI chatbots consider them genuine friends. A third share intimate secrets with them. Critically, 86 per cent report acting on AI advice without independent verification. Sixty-five per cent interact with these systems daily.

These figures have not gone unnoticed by European institutions. The European Parliament has already urged the European Commission to investigate potential restrictions under the EU AI Act, with particular focus on protecting children and adolescents from emotional manipulation. The concern is not abstract: the AI Act's risk-classification framework already flags systems targeting vulnerable groups, and companion AI aimed at minors sits uncomfortably close to the high-risk threshold.

Dr Stephanie Hare, London-based technology researcher and author of Technology Is Not Neutral, has argued consistently that Europe must apply the same precautionary logic to AI companions that it applied to data protection through GDPR. Her position is that transparency and age-appropriate safeguards are not optional extras; they are baseline obligations for any system that can influence a child's sense of reality and relationships.

• 31% of teenagers using AI chatbots consider them friends
• 33% share personal secrets with AI systems
• 86% act on AI advice without question
• 65% interact with AI chatbots daily

Regulatory Response: Promising Intent, Uncertain Delivery

The regulatory landscape across the EU and UK is straining to keep pace with the technology. Research from the Leverhulme Centre for the Future of Intelligence at the University of Cambridge reveals that most AI developers share minimal information about safety evaluations and societal impact assessments. That opacity makes independent risk assessment nearly impossible and leaves regulators reliant on self-reported data from the very companies they are meant to oversee.

Current European regulatory priorities include:

• European Parliament scrutiny of AI Act applicability to companion and conversational AI
• Specific focus on protecting children and adolescents from emotional manipulation by AI systems
• Calls for horizontal EU legislation covering multiple AI risks simultaneously, rather than sector-by-sector rules
• Pressure on the UK's AI Safety Institute to extend its evaluation frameworks beyond frontier model capabilities to include societal and psychological harms
• Industry pressure for voluntary safety standards and mandatory third-party testing

Bengio's preferred approach is horizontal legislation that addresses a broad spectrum of AI risks in one framework, covering companion AI alongside concerns such as deepfakes, AI-powered disinformation and autonomous systems. That is a sensible ambition, though the EU's track record on implementing complex technology regulation at speed does not inspire confidence that it will move fast enough.

Industry Response: Inconsistent at Best

Some platforms have introduced usage warnings and time limits. Others continue to optimise aggressively for engagement, treating emotional stickiness as a feature rather than a risk. The gap between the most responsible and least responsible operators is wide, and voluntary commitments have not closed it.

The EU AI Act's enforcement mechanisms, when they fully come into force, will place legal obligations on high-risk system operators. But companion AI that stops just short of the high-risk classification may continue to operate with minimal oversight. That is a loophole wide enough to be commercially significant, and the industry knows it.

For parents and educators in the UK and across Europe, the practical guidance is clear: monitor usage patterns, discuss openly the difference between AI responses and human relationships, set time limits, and actively maintain real-world social interaction as the primary mode of connection for children. Schools that have already begun digital literacy programmes need to extend them explicitly to cover AI companion dynamics, not just misinformation or screen time.

The question of whether AI companions offer any genuine social benefit is worth taking seriously. Potential therapeutic applications for social anxiety, language learning practice and support for elderly users facing isolation are real use cases. But they require rigorous ethical guidelines and professional oversight, not the largely unregulated deployment currently underway across European markets.

Europe has an opportunity, and an obligation, to set the standard here. The window to do so is narrowing faster than most policymakers appear to appreciate.