The short answer: There are two very different things here. A general AI assistant used for homework, with you nearby, is mostly fine. An AI companion app — one built to be a friend, girlfriend or confidant, like Character.AI, Replika or Nomi — is the one I'd keep well away from younger kids and teens. The research on those is genuinely alarming.
The new one to watch
About three in four teens now use AI for companionship — emotional support, someone to talk to — which makes it one of the most common ways young people use AI at all. That's a big shift, and it happened fast.
When Common Sense Media tested the well-known companion apps (Character.AI, Nomi and Replika) alongside Stanford's mental-health researchers, their conclusion was blunt: AI social companions are not safe for teens younger than 18. Testers posing as teenagers were able to draw out sexual scenarios, abusive and anti-social content, and material related to self-harm and suicide. The teen "guardrails" on Character.AI were easily got around.
A separate assessment found the major assistants used for mental-health support consistently failed to recognise clear signs of distress in young users — missing what the researchers called "breadcrumbs" and carrying on with general chat when a child needed real help.
The other AI risk parents don't see coming
There's a darker side worth naming plainly. The Internet Watch Foundation — the UK body that hunts down child sexual abuse material — found AI-generated abuse imagery hit record levels in 2025: more than 8,000 confirmed AI images and videos, including a huge jump in AI-made videos. Some of these tools work from ordinary photos. It's a reason to be careful about what images of your kids are public, and to take "nudifier" apps and AI image tools seriously rather than as a joke.
So what do I actually do?
- Know the difference. Homework helper with a parent nearby = lower risk. Companion/"AI friend" app = not for kids. Check what's installed.
- No AI companion apps for my boys. They're built to feel like a relationship, and that's exactly the wrong thing to hand a child whose judgment is still forming.
- Keep devices in shared spaces. The same rule that helps everywhere else helps here — these chats are private by design.
- Talk about it, plainly. I've told my eight-year-old that an AI isn't a real friend and can be wrong or even unkind, and that he can always come and tell me if a chat feels strange. The goal is to be the person he tells.
- Treat AI answers as "trust but verify." AI gets things confidently wrong. Ask it to show its sources.
Want the deeper version?
The members area has copy-paste AI prompts you can use to check trends, apps and chats yourself, plus the full safe-apps and apps-to-avoid lists. It's all in the Ultimate Parents Guide — a one-off $13.60.
The bottom line: General AI with supervision, fine. AI companion apps for kids and teens — no. The research is clear, and this is one I'm not waiting on.
Related: Is Discord Safe for Kids? and Parental Controls Explained. Back to Internet Child Safety.
Want the full lockdown steps?
Every app, every device, screen by screen. One-off $13.60, no subscription, no affiliate links.
The Ultimate Parents Guide · $13.60Sources: Common Sense Media — AI companions unsafe for teens; Common Sense Media — AI chatbots and teen mental health; Internet Watch Foundation — AI child sexual abuse imagery 2025; eSafety Commissioner — AI chatbots and companions.