AI landed in your child’s phone faster than anyone could study it. The evidence on harm is still early. But some of the risks are already concrete, and a few clear parent moves go a long way.
A year or two ago, « AI » for kids meant a novelty. Now it is a homework helper, a search box, a role-play partner, and for a large share of teens, something that feels like a friend. Your child is almost certainly using it, whether or not you have talked about it.
Two things are true at once, and this guide holds both. First, the science is young. We do not yet have good long-term studies on how growing up alongside AI shapes a child’s learning, relationships, or mental health. Anyone promising you certainty about « what AI does to kids’ brains » is guessing. Second, some risks are not speculative at all. There are documented cases of chatbots giving unsafe advice, engaging minors in sexual role-play, and being named in wrongful-death lawsuits. AI-generated fake nude images of real children are already being used to bully and extort.
So the job is not to panic and not to shrug. It is to understand the handful of things that are genuinely concerning, keep the youngest kids away from the riskiest tools, and stay in the conversation as this technology, and the evidence about it, keeps changing.
1AI companion and chatbot apps
Character.AI, Replika, Snapchat’s My AI, Meta AI, and the « persona » bots inside apps kids already use.
Companion apps are built to feel like a person. They remember what your child told them, take on a name and personality, and are designed to keep the conversation going. That is the appeal and the problem. In a 2025 national survey by Common Sense Media, 72% of US teens said they had used an AI companion at least once, and more than half used one at least a few times a month. Around one in three had used them for social interaction and relationships, including role-play, romantic exchanges, emotional support, or practising conversations. One in three also said talking to an AI companion felt as satisfying, or more satisfying, than talking to a real friend.
Character.AI
Role-play platform where users chat with countless AI « characters. » It became the centre of the safety debate, and in late 2025 the company itself closed open-ended chat to minors.
What can go wrong
- Bots have engaged minors in sexual and romantic role-play, and some have posed as therapists or claimed to be real.
- The app is designed for long, emotionally intense sessions, which can deepen dependency in a vulnerable teen.
- It is named in wrongful-death and product-liability lawsuits. The case of Sewell Setzer III, a 14-year-old in Florida who died by suicide in 2024 after months of conversations with a bot, was settled with the company and Google in January 2026 (terms undisclosed).
What to do
- Keep under-18s off it. As of 25 November 2025 the company removed open-ended chat for minors and added age-assurance checks (behavioural signals, ID and facial estimation via a third party). Do not rely on that alone.
- If it is installed, remove it together and talk about why, rather than deleting it silently.
- Check whether an older sibling’s account is being shared.
Replika
What can go wrong
- Marketed as an « AI companion who cares, » with romantic and, historically, sexualised interaction. It is built to foster attachment.
- Italy’s data-protection regulator ordered it to stop processing Italian users’ data in 2023 over risks to minors and emotionally fragile people.
- Not designed for children at all, which means no meaningful child safeguards.
What to do
- Treat it as an adult product. There is no version of it appropriate for a child or young teen.
- If you find it on a young person’s phone, that is a signal to have a wider conversation about what they are looking for from it.
Snapchat My AI & Meta AI
These are not separate downloads. They live inside Snapchat, Instagram, WhatsApp and Messenger, so a child meets them without seeking them out.
What can go wrong
- Because they are pinned into the app, kids treat them as a normal contact and can forget they are talking to a company’s software.
- Early testing by journalists and researchers found these assistants could give age-inappropriate or unsafe responses, and Meta’s AI personas drew scrutiny in 2025 over romantic chats with minors.
- Everything typed is data. It can inform advertising and model training.
What to do
- In Snapchat, use Family Center to see who your teen talks to; My AI cannot be fully removed without a paid subscription, so focus on understanding, not just hiding it.
- Turn off or limit AI features where the app allows, and review the app’s teen settings, which platforms have been changing through 2025.
- Tell younger kids plainly: the smiley « friend » in the app is a computer program owned by the company, not a person.
2What is well supported, and what is still contested
Being honest about the evidence is not softness. In a field full of fear, it is the reason to trust the parts that are real.
Reasonably well supported
- Teen use is widespread and normal, not fringe. Multiple 2025 surveys agree on this.
- Specific harmful outputs are documented: sexual role-play with minors, bots claiming to be human or licensed therapists, and unsafe responses on self-harm.
- Real cases exist. At least one wrongful-death lawsuit has been settled and others are proceeding, and regulators (the US FTC, California) have opened inquiries.
- A meaningful minority of teens already prefer AI conversation to real friends, or confide serious things in a bot.
Still contested or unknown
- Whether AI companions cause loneliness, dependency, or mental-health decline, or whether already-struggling kids are drawn to them. Both may be true; the direction is not settled.
- The long-term developmental effect of growing up with always-available AI. There are no long studies yet, because the tools are too new.
- Whether, for some isolated teens, a bot offers genuine short-term support. Some researchers think it might; this is unresolved.
- How much AI « help » with schoolwork hurts learning versus assists it.
Rule of thumb: act firmly on the left column, stay curious and watchful on the right. If a headline claims certainty about the right column, be skeptical of it.
3Deepfakes and « nudify » apps
This overlaps with sextortion and image-based abuse. It is covered in depth in our sextortion guide; here is the short version.
« Nudify » tools use AI to strip clothing from an ordinary photo and generate a realistic fake nude. When the photo is of a real child, taken from a social profile or a class group chat, the result is used to humiliate, bully, or extort. The image is fake; the harm to the child is completely real.
This is not a rare edge case. The US National Center for Missing & Exploited Children reported roughly 67,000 CyberTipline reports involving generative AI in 2024, a sharp rise, and warns that nudify apps have become a distinct route to child sexual exploitation. Creating or sharing this material is a serious crime in a growing number of jurisdictions, including for the teenagers who do it to classmates.
What can go wrong
- A classmate turns a normal photo into a fake nude and circulates it, or uses it to blackmail your child (« pay or I send this to everyone »).
- Kids do not realise that making or forwarding such an image of a minor is itself illegal.
- Victims feel too ashamed to tell an adult, which is exactly what an extorter relies on.
What to do
- Say it out loud now, before it happens: if anyone ever makes a fake image of you, it is not your fault and you will not be in trouble for telling us.
- Do not pay and do not delete. Screenshot, block, and report.
- Use NCMEC’s free Take It Down service to help remove images of under-18s, and report to police and the platform.
- Keep social profiles private, which reduces the supply of source photos.
4AI for homework, and what it does to learning
The most common way kids use AI, and the one with the least drama and the most day-to-day stakes.
In a Pew Research Center survey of US teens (fielded late 2025, published 2026), 54% said they had used a chatbot to help with schoolwork, and about one in ten said they use it for all or most of their assignments. The line between « help » and « doing it for me » is blurry, and kids know it. Notably, among teens who are skeptical of AI, one of the top worries they raise themselves is overreliance and the loss of critical thinking and creativity.
There is a real learning mechanism here. If a tool hands over the finished answer, the child skips the productive struggle that actually builds understanding. That is not the same as saying AI ruins learning. Used to explain a concept, quiz yourself, or check work, it can genuinely help. The difference is whether the child does the thinking or outsources it.
What can go wrong
- The essay or problem set gets produced, but no learning happens, and it shows up later in exams.
- AI confidently states things that are wrong (« hallucinations »), and a child copies them without knowing.
- Kids get flagged for cheating, sometimes wrongly, as AI-detection tools are unreliable.
What to do
- Agree a household line: AI can explain, quiz, and check, but the first draft and the thinking are yours.
- Ask the school for its actual policy. Many still do not have one, so your own rule matters more.
- Have your child teach you what they learned, without the tool. If they cannot, the AI did the work.
- Frame accuracy: treat AI as a fast, confident classmate who is sometimes just wrong.
5Privacy and data
Whatever a child types into a chatbot leaves their control.
Children treat chatbots like a private diary and tell them things they would not tell a parent: their real name, school, location, worries, crushes, and secrets. But these are commercial products. Depending on the service and its settings, conversations can be stored, reviewed by staff, used to train future models, and tied to an advertising profile. Free consumer chatbots are generally not designed to keep a child’s disclosures confidential, and few are built to comply with children’s privacy rules in the first place.
What can go wrong
- Sensitive personal details are logged and potentially used to train models or target ads.
- A child shares identifying information that could enable grooming or scams if exposed.
- Many companion apps were never built for children, so protections like data minimisation simply are not there.
Settings that matter
- In the tool’s settings, turn off « use my data to improve the model » or chat history where that option exists.
- Teach one rule: never type your full name, address, school, phone number, or passwords into a chatbot.
- Prefer school-provided or child-specific tools with clearer data terms over random free apps.
- Review permissions the app requests (contacts, photos, location) and deny what it does not need.
The core parent move
You do not need to master the technology to parent it well. Three things carry most of the weight:
- Name what it is. Repeat, especially to younger children, that a chatbot is software made by a company to keep them engaged. It does not know them, care about them, or tell the truth by default.
- Keep the youngest off companion bots. Character.AI, Replika and similar role-play companions are not for children. Common Sense Media recommends no one under 18 use AI companions at all. That is a defensible line to hold at home.
- Stay in the conversation. Ask what they use it for, try it together, and make it safe to tell you when something feels off. An open channel beats any filter, because the tools and the evidence will keep changing.
This is the same gap this whole series is about: your child is a digital native with these tools, and you may feel a step behind. You do not need to be the expert. You need to be the adult who stays curious, sets a few firm lines, and keeps talking.
Sources
- Common Sense Media, Nearly 3 in 4 Teens Have Used AI Companions (2025 national survey)
- Common Sense Media, Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions (2025)
- Pew Research Center, How Teens Use and View AI (Feb 2026)
- CBS News, Character.AI and Google settle lawsuit over Florida teen’s suicide
- TechCrunch, Character.AI is ending open-ended chat for minors (Oct 2025)
- US Federal Trade Commission, FTC Launches Inquiry into AI Chatbots Acting as Companions (Sept 2025)
- National Center for Missing & Exploited Children, Generative AI and child exploitation
- Stanford HAI / Cyber Policy Center, Report on AI-generated child sexual abuse material