AI chatbot bans for kids: what the 2026 laws actually do

Somewhere in your house tonight, a teenager may be telling a chatbot something they have not told you. That gap, between what a kid says to an app and what they say to a parent, is what lawmakers spent much of 2026 trying to close.

The tools they reached for are blunt. Several US states have passed laws aimed squarely at AI “companion” chatbots, and a federal bill is working its way through the Senate. If you have only half-followed the headlines, here is what is actually on the books, and where it stops.

What the new laws actually require

California’s SB 243 took effect on 1 January 2026, and it reads like the template everyone else is copying. Under the law, a companion chatbot has to tell a user that its replies are generated by AI and not a human. It has to nudge minors to take a break at least every three hours of continuous use. Operators must run protocols to head off content about suicide, self-harm and suicidal ideation, and take reasonable steps to stop a chatbot serving sexual material to a minor. From 1 July 2027 they will also have to report flagged cases of suicidal ideation to the state, and families can sue when an operator’s failure causes harm.

The federal bill goes further. The GUARD Act, introduced in October 2025 by senators Josh Hawley and Richard Blumenthal, has cleared the Senate Judiciary Committee. It would require real age verification before someone can use a companion chatbot, and, tellingly, it says letting a user simply type in a birth date does not count. It bans chatbots that encourage suicide, self-harm or violence, or that draw minors into sexual conversations, with fines that can reach into six figures for each violation.

(A disclosure, since this is our patch: Mentus AI makes AI mentoring tools for children, so we have a stake in how these rules are drawn. The bills generally carve out educational tools from the harshest restrictions. We still think the safety bar should sit high, and we would rather be measured against it than around it.)

The trouble with the word “companion”

Here is the catch. These laws go after a category, the “companion” chatbot, that is harder to pin down than it sounds.

When Common Sense Media surveyed 1,060 US teens in 2025, it found that 72% had tried an AI companion at least once and 52% used one regularly. But the researchers counted more than the obvious apps like Character.AI or Replika. They also included general-purpose chatbots such as ChatGPT, when a teen uses them for personal, companion-style conversation. And a child does not switch products when the chat drifts from a history essay to how they are feeling. The tool stays the same. Only the conversation changes.

That is the soft spot in any ban. A law can wall off an app that markets itself as an AI friend. It is much harder to police the moment a homework helper quietly becomes a confidant.

Not everyone thinks a ban is the fix

Some of the people closest to this worry that the blunt instrument misses. Writing in STAT in July 2026, RAND and Harvard researcher Ryan McBain argued for rules that sort chatbot conversations by risk rather than banning them outright: ordinary chat, clinical-style advice, and outright crisis each call for a different response. He pointed out that the share of young people turning to AI for mental-health advice rose more than 40% in a single year, citing JAMA Pediatrics figures, at a time when, according to CDC data, fewer than half of teenagers with a major depressive episode got any counselling in 2023.

His point is an uncomfortable one. For some kids, a flawed chatbot is the only thing that answers at two in the morning. Ban it clumsily and you may pull away a crutch without putting anything better in its place.

Our own read is that both things are true at once. The 2026 laws are a fair response to real harm, and they will not do a parent’s job.

What to do while the law catches up

None of this lands in your living room this year. Age verification is still being fought over, the reporting duties do not start until 2027, and the federal bill is not law yet. So the useful move is smaller and sooner than waiting for Washington.

Find out which kind of chatbot your child actually talks to. Open it yourself. Is it an app built to be a friend, or a general assistant they have started confiding in? Then ask them, without turning it into a raid, what they use it for. The research is consistent on one point: kids are having these conversations whether or not the adults have noticed. The parents least rattled by the headlines tend to be the ones who already know what is on the screen.