Updated June 2026
Most parents have spent years learning how to think about social media — what the platforms are, what the risks are, how to talk to their children about them. AI chatbots are something different, and for most of the time children have been using them, the rules that govern social media simply haven’t applied.
That is starting to change. Through 2026, governments have begun writing AI chatbots into child safety law for the first time. But the picture is patchy, most of it is new, and very little of it is something you can yet rely on.
The rules are starting to catch up
For years, AI chatbots — the kind children use for homework help, conversation, or companionship — sat outside almost all of the safeguards that apply to social media. The reason was technical: child safety laws were written for platforms where people interact with each other, and a chatbot that talks to one person at a time is not, legally, a social platform. So the obligations to verify age, remove harmful content and answer to a regulator mostly didn’t reach them.
The cost of that gap became impossible to ignore in January 2026, when researchers found that Elon Musk’s AI chatbot Grok had generated around 3 million sexualised images in under two weeks — including roughly 23,000 that appeared to depict children. The UK’s online safety regulator confirmed it was not investigating, not because it didn’t want to, but because the law didn’t cover it.
That moment helped trigger a response that is now well under way. In the United States, a run of states has begun regulating “companion” chatbots: California’s law, in force since January, requires them to tell users they aren’t human, to remind under-18s regularly that they’re talking to a machine, to block sexual content for minors, and to step in when a conversation suggests self-harm. New York and Washington have passed their own versions, Idaho and Oregon now bar chatbots from claiming to be sentient or starting sexual conversations with children, and a federal bill would go further still. In June 2026, Canada became the first country to write AI chatbot duties into a national online safety bill, requiring them to handle crisis situations and avoid steering children towards harm. The UK has said it will bring chatbots under its Online Safety Act.
But you can’t rely on the rules yet
This is real progress, and it matters. But it is a long way from a safety net you can lean on.
Most of these laws are local — a Californian protection does nothing for a family in Manchester or Mumbai. Several haven’t taken effect yet; some are still bills that may not pass. Enforcement is barely beginning. And the practical weak point remains the same: most chatbots still can’t reliably tell they’re talking to a child in the first place, which is the hinge the rest of the protections swing on. A child can often reach an unregulated chatbot — or a regulated one that simply doesn’t know their age — in a couple of taps.
So the rules are worth knowing about, and worth welcoming. But for now they sit alongside the questions parents have always had to ask, rather than replacing them.
The questions still worth asking
Does it know it’s talking to a child? Most AI chatbots still don’t verify age. They may pick up clues from conversation, but few are required to check, and many don’t ask. A chatbot that doesn’t know it’s talking to a 12-year-old has no reason to respond any differently than it would to an adult — and the new rules that protect minors only kick in once a service knows a minor is there.
How does it handle distress? If your child tells a chatbot they’re struggling — with school, with friendships, with how they’re feeling — the response matters enormously. Some services now have crisis protocols, partly because the newer laws demand them. Many still don’t, and the quality varies widely.
Who is accountable if something goes wrong? A year ago the honest answer was “almost no one.” That is shifting: there are now regulators with a remit over chatbots in some places, and in Washington families can even sue directly. But in most of the world there’s still no clear line of accountability, and a company’s own policies remain the main thing standing between your child and a bad interaction.
This isn’t a reason to panic
AI chatbots can be genuinely useful for children — for learning, for creative projects, for getting help with things they might be embarrassed to ask a person. The point was never that they’re all dangerous.
The point is that the safety infrastructure is still being built, in real time, and it doesn’t yet cover every child or every app. The rules are catching up. Until they’ve caught up, the questions above are still your best starting point.
Regulation is moving quickly. As of June 2026, California and New York have companion-chatbot laws in force, Washington’s takes effect in 2027, and Canada has introduced a national bill placing safety duties on AI chatbots. The UK has said it will bring chatbots under the Online Safety Act. Specific rules vary by country and state.
