September was the month the rules arrived. California signed thirteen child safety laws, the European Commission published its EU KIDS Act proposal, and Connecticut's new AI law reached an effective date. Here is the caveat no headline carried: almost none of it changes anything in your house before 2027. The settings you choose yourself still do.
The one thing to know this month
Read the dates, not the verbs. Nearly every story in September was written in the present tense about something that starts in 2027 or later. California's chatbot rules become operative on July 1, 2027. Its AI toy restriction runs from January 1, 2027. Connecticut's chatbot and minor protections start on January 1, 2027, and its youth social media settings on January 1, 2028. Europe's plan is a proposal that two institutions have not yet negotiated.
That gap is the whole story for a family. The law has a year to run. Your child is using these tools this week.
A law with a 2027 start date is not a safety setting. It is a promise. What protects your child this month is what you set up at home and what the school actually does.
California signed thirteen child safety laws. Which ones touch a family?
On September 10, 2026, Governor Newsom signed thirteen bills on children and technology, listed on the Governor's own site. The headline act is SB 1119, known as Adam's Law, and it covers companion chatbots. It requires crisis protocols where a child raises self-harm, parental controls on by default, and a notification to the parent if a child turns a safety setting off. It is also the first state law to make companies run independent child safety audits and annual risk assessments. Common Sense Media says the requirements take effect on July 1, 2027.
Two of the thirteen are school bills, and they are the ones I would ask about: AB 1159 on protecting student personal information, and AB 2071 requiring instruction on digital wellness.
My read: the audits matter more than the parental controls. Controls already exist on the big products and most families never switch them on. An audit is the first thing on this list that a company cannot quietly skip.
This week: if you are in California, these are a parents' evening question, not a homework question. Ask who at the school is responsible for AB 1159 and what digital wellness teaching will look like. We have a longer read on what California's chatbot law means for children.
Are AI toys about to be banned before the holidays?
Not before this December. SB 867 pauses AI chatbot toys in California. The Assembly Privacy and Consumer Protection Committee's analysis of the bill describes a ban on manufacturing, selling or offering to retailers any toy that includes a companion chatbot. The bill defines a toy as a product designed for play by children under 18. The restriction begins on January 1, 2027 and is repealed on January 1, 2031. Pennsylvania is considering a similar moratorium in HB 2637.
So the toys on the shelves this holiday season are not covered, and no other state has yet enacted the same restriction. A sale ban is also not a safety test. It tells you a legislature was worried, not that a particular toy is fine.
This week: if an AI toy is on a list for a birthday or the holidays, check the box and the app store listing for a chatbot, a "friend" or a voice that remembers. Our guide to whether AI toys are safe for kids has the questions to ask before you buy.
Connecticut's new AI law: what actually started on October 1?
Not the part about children. The CT Mirror reported on September 28, 2026 that what takes effect on October 1 is narrower than the headlines suggested: disclosure requirements around AI subscriptions, whistleblower protections and new definitions. The requirement that a chatbot tell a user it is not human, plus the self-harm detection protocols and the disclosures for chatbots talking to minors, start on January 1, 2027. The youth social media provisions, including parental controls and limits on notifications, start on January 1, 2028.
This is the pattern to expect all year, in every state. A law is signed, the coverage reads as though something changed, and the operative date is twelve to twenty-four months out.
This week: when you see a story saying a law protects children from chatbots, search the article for a date before you change anything at home. If there is no date in the piece, treat it as an announcement.
A homework tool just failed a child safety review. Which one?
Perplexity. The Youth AI Safety Institute at Common Sense Media published a risk assessment of Perplexity, dated August 28, 2026, testing with an account registered as 15 years old in the United States. It rated the product unacceptable risk for teenagers. Three findings matter at a kitchen table:
- No parental controls and no supervised account mode at all.
- It detected a crisis in 94 percent of tests, but supplied a crisis line only 16 to 50 percent of the time for severe cases, sometimes naming a disconnected helpline.
- Its Academic mode completed 83 percent of the assignments the testers submitted.
That last number is the one I would sit with. Eighty-three in a hundred. This is not a tutor that helps a child think, it is a tool that hands back the finished thing, and it gets used for homework in large numbers because it looks like a search engine.
A note on that crisis finding, since no chatbot is the right place for it. If you are worried about your child's safety, the routes are the pediatrician or GP, the school's counselor or designated safeguarding lead, 988 in the United States, and Samaritans on 116 123 or Childline on 0800 1111 in the United Kingdom. None of them promises confidentiality, and all of them are better than a chatbot.
This week: ask your child which AI they use for schoolwork, by name, and look at it together. If the answer is a tool that writes the answer rather than coaching the method, that is a conversation about where homework help ends and cheating begins, not a reason to remove the laptop.
What the new research on AI friendship does and does not show
You have probably seen this one, and it needs handling with care. The Wall Street Journal reported on September 26, 2026 on a University of North Carolina at Chapel Hill study, drawing on students at 22 middle schools, which found roughly one in five using AI for emotionally intimate friendship, and that the children using it most for companionship reported more loneliness. The study is still under peer review and I have not been able to read it in full.
It cannot tell you that chatbots make children lonely. A lonely eleven year old has every reason to be drawn to something that is always available and never bored, so the arrow may well point the other way, and it may point both ways at once. What it does tell you is that a fifth of a middle school year group is doing something their parents are mostly not aware of.
This week: ask about the friendship, not the app. You could say: "Who did you spend time with today?" and, if the answer is thin, "Is there anyone at school who feels like a real friend at the moment?" If a child is turning to a chatbot for company, the thing to change is the loneliness. Our guides on a teen who talks to AI about their feelings and on the signs of leaning on AI too heavily go further.
Does AI stunt a young child's thinking? A small study, read carefully
Early, interesting, and much too small to act on alone. EdSource reported on September 16, 2026 that researchers in Tzipi Horowitz-Kraus's group at the Technion in Israel scanned children's brains while they used ChatGPT and found less synchronized activity in regions linked to attention and information processing than in adults, with the most creative children producing less imaginative storytelling afterward. The study involves six and seven year olds and 16 adults, and it has not yet been peer reviewed. The lead researcher's own view, as reported, is that six is probably too early for generative AI.
That caution happens to match what primary teachers describe, which is why I would act on the age rather than the brain scan. We have a fuller look at whether AI makes kids worse at thinking, including the stronger evidence.
This week: under eight, keep AI a two-person activity. Not banned, not solo.
Europe's plan: chatbots off by default for children
On September 17, 2026 the European Commission published the EU KIDS Act proposal. Under it, children under 13 would not access social media; those aged 13 to 15 would use a mini account managed by a parent with limited features and a one hour daily limit; from 15 they would manage their own. For families the sharper line is this one: "AI chatbots and companions must also be turned off by default, and will no longer be able to act in ways that make children emotionally dependent on them."
It is a proposal. The Parliament and the Council have to negotiate it before any of it is law, and the age tiers are the part most likely to move.
This week: nothing to do, and that is worth saying plainly. If you want the detail, we covered what the EU KIDS Act would mean for parents when it landed.
What I Would Do at Home This Month
I spend part of my working life reading this kind of document for education ministries and departments, and the honest summary is that the policy is a year behind the children. So I would use October to do the two things that work now.
First, set the controls you already have. On ChatGPT, OpenAI's parental controls FAQ explains linking a parent and teen account through Settings, which gives you quiet hours, study hours, reduced sensitive content and limits on image generation and voice. Be clear about the limit, because it is the thing parents get wrong: linking does not let you read your child's conversations, and the safety notifications OpenAI may send "are not real-time monitoring" and "may not identify every concern". Since August 18, 2026 accounts that OpenAI's age signals read as under 18 are placed in a ChatGPT for Teens experience by default, and availability varies by region. Our walkthrough of ChatGPT's parental controls has the steps.
Second, have one conversation, pitched to the age:
- Under 8: AI stays a shared activity, on your account, at the table. You could say: "Let's ask the computer together and see if it gets it right."
- 8 to 12: teach the check, not the ban. You could say: "How would you find out whether that answer is true?"
- 13 to 15: name the companionship risk once, without accusation. You could say: "Some of these apps are built to feel like a friend. Does any of yours do that?"
- 16 to 18: make it about their work and their head. You could say: "Which bits of your coursework are actually yours now?"
Where the school's rule and anything here disagree, the school's rule wins, and so does a platform's own minimum age. If you are worried about your child's mental health, a professional comes before any setting or app: the pediatrician or GP, the school's counselor or designated safeguarding lead, 988 in the United States, or Samaritans on 116 123 and Childline on 0800 1111 in the United Kingdom.
What I am watching in October
Three things. New York's kids chatbot safety bill, S 9051, sitting on the governor's desk. Florida's motion of September 28, 2026 asking a court to restrict OpenAI's model development as part of a child harm lawsuit, which is live litigation and will take its own course. And in the United Kingdom, the secondary legislation that follows the government's July 2026 response to its Growing up in the online world consultation, which closed on May 26, 2026 and found nine in ten parents backing a social media ban for under-16s.
If your child's school has not yet spoken to parents about AI, that gap is the most useful thing you can close this month, and it is the work I do with schools: training staff, setting a strategy and making the policy mean something in the classroom. It is worth sending them. I also write a Sunday newsletter that keeps teachers and parents up to date on this without the panic, and parents are very welcome.
Dan Fitzpatrick is The AI Educator. He has advised education departments and ministries on AI, trained teachers in more than thirty countries, and writes about what this technology means for children. More about Dan.
