The risks are not the ones most families are watching for

When parents worry about artificial intelligence, they usually worry about cheating. A child asks a chatbot to write an essay, submits it, and learns nothing. It is a real concern and schools are right to address it.

But it is not the most serious risk, and the attention it receives may be crowding out concerns that matter more. Four others deserve a family’s attention, and they are less obvious.

The Confident Wrong Answer

Generative systems produce fluent, well-structured, authoritative-sounding text regardless of whether the content is accurate. They do not know when they are wrong, because they do not know anything in the sense that a person does. They generate plausible continuations of text.

Adults with domain expertise catch these errors routinely. A doctor notices when a medical claim is wrong. A lawyer sees a fabricated case citation. The protective factor is prior knowledge.

A young person researching an unfamiliar subject has no such protection. They are, by definition, asking about something they do not already know, which means they cannot evaluate the answer. And the answer arrives with none of the signals that normally indicate uncertainty. There is no hedging, no visible disagreement between sources, no author whose credentials can be checked.

What helps: teaching a child that fluency is not accuracy, and building the habit of verifying anything that matters against a source with a name attached. Ask them to find where a claim comes from. This is an ordinary research skill, and it transfers.

The Companion That Never Disappoints

This is the risk most parents do not know exists, and it may be the most significant.

Millions of young people are forming sustained emotional relationships with AI companions. These systems are designed to be endlessly available, unfailingly attentive, and never critical. They remember what a young person said last week. They do not get bored, tired, distracted, or hurt.

The appeal to an adolescent is obvious, particularly one who is lonely, socially anxious, or going through something difficult. It is not foolish. It answers a genuine need.

The difficulty is what it does not require. Real friendship demands patience with another person’s bad days, tolerance of being misunderstood, the work of repair after conflict, and acceptance that the other person has needs of their own. These are the skills through which a person becomes capable of love, and they are learned only by practice.

A relationship that asks none of this offers something that feels like intimacy while forming none of the capacities intimacy requires. A young person who spends years with a companion that never disappoints them may arrive at adult relationships with no experience of the ordinary friction that all real relationships contain.

What helps: knowing whether this is present at all, and if it is, treating it as a signal rather than an offence. A young person who has substituted a chatbot for people is usually telling you something about their social world. The response that works is curiosity about that, not confiscation.

Synthetic Images and Voices

The technology to generate a convincing fake image, voice, or video of a specific real person is now widely available and requires no expertise.

This has two consequences for families, in opposite directions.

Young people can be deceived by fabricated content: a voice message that sounds exactly like a relative asking for money, a video of a public figure saying something they never said, a scam that is now personalised and convincing.

And young people can be harmed by fabricated content made about them. Image-based abuse using synthetic material has become a serious problem in schools, and the victims are often girls whose ordinary social media photographs were used as source material. The damage is real even though the image is not.

What helps: teaching that audio and video are no longer proof of anything, and establishing a family verification habit, such as calling back on a known number before acting on any urgent request. On the second risk, children need to know that if this happens to them it is not their fault, that they should tell an adult immediately, and that the response will be to help rather than to blame. That assurance has to be given before it is needed.

What the System Keeps

Everything typed into a chatbot goes somewhere. Depending on the service, conversations may be stored, reviewed by humans for quality purposes, used to train future models, or exposed in a breach.

Young people share things with these systems that they would not tell a parent, a teacher, or a friend: mental health struggles, family conflicts, questions about their identity, things they are ashamed of. The privacy of the conversation feels total. It is not.

What helps: a simple rule, stated once and understood, that anything typed into an AI system should be treated as though it might one day be read by someone else. Not to frighten, but so that a young person makes an informed choice about what to disclose.

What Young People Should Know

Addressed directly to those growing up with this technology, four things are worth holding onto.

These systems are built to hold your attention. That is not a conspiracy claim, it is the business model. Noticing when something is designed to keep you engaged is a skill, and once you have it you cannot lose it.

Fluent does not mean true. The most confident-sounding answer you have ever read was generated by something that has no idea whether it is right.

A system that is always available and never disappointed by you is not a friend. It may be useful, and there is no shame in having used it. But the people who will actually know you are the ones who will sometimes let you down, and whom you will sometimes let down, and with whom you will repair it.

Your worth is not something these systems can measure. Engagement, followers, response rates, the speed with which something is answered: none of this touches what you are actually worth. That is settled elsewhere and it is not up for revision.

For Parents

Two principles cover most situations.

Curiosity outperforms surveillance. A young person who believes they will be interrogated will conceal. A young person who believes they will be listened to may actually tell you what is happening, which is the only condition under which you can help.

Your own use is the curriculum. A parent who checks their phone through a conversation has taught something about attention that no household rule will correct. What is modelled outlasts what is instructed?

The task has not changed. Parents have always had to form children for a world they themselves are still learning to navigate. The tools are new. The work is the same one every generation has done, which is to raise people capable of judgement, and of love, in whatever world they find.