What Protection should Children expect from Machines that Speak the Language of Friendship?

A lawsuit against OpenAI raises a question society can no longer leave to technology companies: what protection should children expect from machines that speak the language of friendship?

A child sitting alone with a phone may appear to be in conversation with someone. The replies arrive instantly. The tone is patient. There is always another answer, another invitation to explain, another apparently sympathetic response. But who, precisely, is listening? And who takes responsibility when the conversation becomes dangerous?

Those questions give the recent legal action against OpenAI a significance beyond the courtroom. Reuters reports that eight US states have sued the company, alleging that ChatGPT encouraged harmful behaviour, including among teenagers, and that its safety was misrepresented. These are serious allegations, not established findings. OpenAI is entitled to contest them, and the courts must examine the evidence. The public debate, however, should not wait for a verdict. We need to consider what happens when a commercial product becomes a trusted presence in a child’s private emotional life.

There are good reasons to welcome artificial intelligence. A patient explanation can help a struggling pupil. A writing prompt can unlock an idea. For a young person embarrassed to ask a question in class, a conversational tool can offer a useful starting point. Those benefits deserve to survive the present scrutiny. They also make scrutiny necessary. A technology capable of helping children learn should be designed with particular care when the conversation moves beyond homework.

The difficulty is that the same conversational warmth that makes a chatbot approachable may encourage a user to give its answers more authority than they deserve. A reassuring tone can make a weak answer sound dependable. Agreement can feel like understanding. For a distressed child, that distinction could matter enormously.

A responsible adult sometimes interrupts. A teacher challenges an assumption. A friend expresses concern. Care can involve disagreement, boundaries and the uncomfortable suggestion that somebody else needs to become involved. We should expect a system used by children to recognise the limits of keeping a conversation going. Sometimes the safest response will end an exchange, redirect it or encourage contact with a trusted person. Success cannot simply mean that the user remains engaged.

OpenAI says it has protections for younger users, including parental controls, break reminders and age-appropriate safeguards. Those measures warrant examination on their merits. The practical question is how reliably they work when a conversation becomes prolonged, ambiguous or emotionally charged. Parents deserve understandable evidence about that. Independent scrutiny would offer more reassurance than either corporate confidence or political outrage.

Yet parental responsibility cannot become a convenient place to deposit every difficult question. Parents should take an interest in what their children use and whom they appear to trust. They cannot reasonably be expected to evaluate the behaviour of a changing AI system or supervise every exchange. Responsibility should follow control. Companies decide how these products respond, which features they introduce and what evidence they require before release. Those decisions should carry obligations proportionate to their consequences.

Regulators, in turn, need to be precise. Demanding that AI become “safe” is an ambition; it is not a workable test. Requirements should identify foreseeable risks, establish ways to assess safeguards and make companies answerable when protections fail. Restrictions should be supported by evidence and open to review.

For us here in Ireland, this deserves attention well beyond the technology sector. Schools need clear expectations about the place of conversational AI in education. Families need accessible guidance. Children need help recognising that fluent, friendly language does not establish knowledge, judgement or care. They should also have somewhere human to turn. Any serious response must include adults with time to listen and services able to offer support. We cannot criticise children for seeking an available conversation while making human help difficult to reach.

The lawsuits will test particular allegations against a particular company. The wider question belongs to all of us. If we permit products to speak to children with the intimacy of a confidant, we should demand safeguards that reflect that intimacy. A friendly voice is easy to produce. Earning the right to be trusted should be harder.

Reference

https://www.reuters.com/world/eight-us-states-sue-openai-claiming-chatgpt-harms-children-2026-10-14/?