Project report
Written Evidence: Toward Child-Centred AI Safety: Artificial Intelligence Harms & Children’s Online Safety
Gina Neff, JL Freeman ·
Summary
Written by AI, not yet AICS's view · claude-haiku-4-5-20251001This evidence submission to the UK Parliament examines how artificial intelligence poses risks and opportunities for children's online safety. It identifies emerging harms including AI-generated misinformation, manipulative chatbots, non-consensual deepfakes, and parasocial relationships with AI, with girls facing disproportionate risks. The authors call for child-centred approaches that include children's voices in policy design rather than blanket bans, and emphasise that regulation, media literacy, and data transparency are all necessary together.
Key points
- AI is embedded in children's online environments but existing safety frameworks were not designed for children's digital practices
- Key emerging harms include health misinformation, emotionally manipulative chatbots, non-consensual intimate deepfakes, and psychological risks from AI companions
- Girls and young women face disproportionate risks from AI-related harms
- Children recognise real benefits from AI in education, healthcare access, and social support
- The UK Online Safety Act has significant gaps in coverage of AI chatbots and deepfakes; regulation, media literacy, and data transparency must work together
Limitations
This is a policy evidence submission based on literature review and existing research rather than new empirical data collection. It does not evaluate the effectiveness of any specific intervention through original studies or randomised trials.
Abstract
This written evidence submission to the UK All-Party Parliamentary Group (APPG) on Children's Online Safety examines the risks and opportunities that artificial intelligence (AI) poses for children and young people in the United Kingdom. Authored by Dr. Jaimie Lee Freeman and Prof. Gina Neff of the Minderoo Centre for Technology and Democracy at the University of Cambridge, the submission draws on original and published research to address three of the inquiry's four questions. AI is now deeply embedded in children's online environments—across social media platforms, search engines, and video games—yet existing safety frameworks were not designed with children's distinct digital practices in mind. This submission identifies key emerging harms, including AI-generated health misinformation, sycophantic and emotionally manipulative chatbot interactions, non-consensual intimate deepfakes, and the psychological risks of parasocial relationships with AI companions. These harms fall unevenly across communities, with girls and young women facing disproportionate risks. Despite these concerns, children themselves recognise real benefits from AI in education, healthcare access, and social support. The evidence strongly cautions against blanket bans, which evidence from Australia suggests can obscure rather than resolve the underlying harms young people experience online. This submission calls for decision makers to consider child-centred AI safety approaches that reflect children's developmental diversity, embed their voices in participatory research and policy design, and go beyond harm mitigation to maximise AI's benefits for future generations. On regulation, the authors identify significant gaps in the UK Online Safety Act's coverage of AI chatbots and deepfakes and argue that meaningful progress requires improved data transparency from industry, expanded researcher access to platform data, and robust auditing frameworks. Media literacy education, regulation, and data transparency are each necessary but individually insufficient; only in combination, and grounded in children's lived experiences, can they deliver the safe digital futures children deserve.