Researchers Urge a Precautionary Ban on Children's Use of AI Chatbots

A newly published paper argues that children should not be permitted to use large-language model chatbots such as ChatGPT, Gemini, and Claude until their safety can be demonstrated. The authors frame this as a precautionary measure intended to protect young people from serious and potentially irreversible harm.

The paper draws a parallel with social media. After two decades of largely unregulated use, a substantial body of research now documents adverse effects on children, prompting responses such as Australia's ban on social media for those under 16. The authors contend that society waited for overwhelming evidence before acting, and that a precautionary stance would have prevented harm.

They apply the precautionary principle, which they say requires four conditions: uncertainty about the risks, a credible scientific basis for believing serious harm could occur, the prospect of irreversible damage, and a proportionate response. Central to their argument is shifting the burden of proof onto those who would allow access, requiring them to show that benefits outweigh risks, much as medicines are assessed before market approval.

The authors point to early empirical findings and high-profile incidents, including deaths, that suggest chatbots can contribute to:

• emotional dependency
• impaired problem-solving skills
• worsening mental-health vulnerabilities

They acknowledge that some studies report modest benefits, but describe the current evidence base as preliminary and inconclusive. Given the plausibility of severe harm and the absence of robust safeguards, they conclude that a temporary, proportionate ban is the most prudent policy, lasting until developers can prove the net benefits outweigh the risks.

The original research was reported in AI and Ethics (Springer Nature).