Norton released a survey this morning on children, parents, and artificial intelligence. The number the press release leads with is that 85 percent of American parents feel comfortable talking to their kids about online risk. The number that should stop you sits four bullets further down.
Thirty-eight percent of parents whose child plays a chat-enabled game say someone asked that child to move the conversation to a different platform.
That is not a new AI risk. That is the oldest grooming step there is, and it has worked the same way since AOL Instant Messenger. What has changed is that the person on the other end may now be running a script.
What The Study Reports
The Norton Insights Report: Connected Kids surveyed 1,000 US adults between May 22 and June 9, 2026. Norton is a brand of Gen Digital, the company formed when NortonLifeLock and Avast combined. Among US parents of children under 18, the study found:
- 32 percent are not confident their child can tell AI-generated content from real content
- 67 percent worry AI could be used to build fake images or conversations involving their child
- 58 percent have already restricted apps, monitored activity, or talked to their child about AI
- 40 percent say their child uses a device past the hour they set
- 22 percent say their child reached a platform the parent believed was blocked
- 20 percent say their child shared private details with a stranger
“Parents have spent years teaching children how to recognize strangers online, but AI changes the nature of what ‘danger’ online looks like,” said Leyla Bilge, Global Head of Scam Research at Norton. “From the misuse and alteration of posted or shared photos to AI bots acting as cyberbullies or impersonating trusted sources, there are entirely new risks at play.”
Alongside the study, Norton launched The Bots & The Bees, a set of guides on sharenting, stranger danger in an AI era, and teaching kids to question what they see.
Parents Are Guarding The Wrong Door
Ask parents which platform worries them and they name TikTok, at 21 percent. Facebook comes second at 13 percent. YouTube is third. Online gaming lands fourth, at 10 percent.
Now look at what parents report actually happening. Among the 64 percent whose child plays chat-enabled games, 64 percent say the child has talked with a stranger by voice or text, 36 percent say the child received offensive or inappropriate messages, and 38 percent say the child was pushed to continue the conversation elsewhere.
The platform ranked fourth for concern is producing the highest reported exposure by a wide margin. Bilge named the reason plainly: the chat layer sitting on top of a game is less visible to a parent than a social feed. Nobody scrolls their kid’s voice channel.
Every One Of These Numbers Is A Floor
Read the methodology again. Norton surveyed adults, not children. Every finding here measures what a parent found out about, not what happened to the child.
So when 20 percent of parents say their child gave personal details to a stranger, the honest reading is that 20 percent of parents caught it. The 22 percent who discovered their kid slipped past a block are the 22 percent whose kid got sloppy. A twelve-year-old who successfully hides a Discord account does not show up anywhere in this data.
Treat every percentage in this report as the low end of a range. I have never seen a proxy-report survey on adolescent behavior where the parent number came in higher than the kid number.
Norton Undersold Its Own Source
The release states that the National Center for Missing & Exploited Children received “more than 100 reports every day” of financial sextortion last year, and links to the NCMEC page as support.
Open that link. NCMEC’s own June 2026 announcement reports more than 50,000 financially motivated sextortion reports in 2025. That works out to 137 per day, up from roughly 36,000 reports the year before. A 39 percent year-over-year jump.
“More than 100” is technically accurate. It also shaves better than a third off the figure NCMEC published, in a release arguing that the threat is growing. I doubt anyone did that on purpose. Somebody wrote the sentence against 2024 data and never went back after the new report landed. It is worth flagging anyway, because that “100 a day” line is the one every outlet will quote by Thursday, and the real number is worse.
The same softening shows up in the AI video statistic. The release says “over 3 percent” of AI-generated videos are scams. Gen’s H1 2026 threat report puts the monthly average at 3.2 percent across tens of millions of video encounters. That tracks with what federal investigators have been saying about AI-driven fraud fooling Americans at scale.
How I Read A Vendor Study, And Why You Should Too
I have spent 25 years reading research published by companies that sell something. Most of it is useful. The tell is almost never a wrong number. The tell is a number with no base attached to it.
This release says 1,000 US adults were surveyed, then reports finding after finding as a percentage “of parents.” Those are two different denominators, and the release never gives you the second one.
Norton’s full report does. Parents of children under 18 made up 39 percent of the adults surveyed. Run the arithmetic on the US sample and you get roughly 390 parents, not 1,000. That is a respectable base for a headline finding.
The subgroups are where it gets thin. Roughly 250 of those parents have a child in chat-enabled games. In the US, about 25 percent report a cyberbullying incident, which is somewhere near 98 parents. The finding that 16 percent of those named an AI bot or AI-generated account as the perpetrator rests on something like sixteen people.
Sixteen respondents is a signal. It is not a statistic you build a policy on, and it is going to be repeated for the next year as though it were. That arithmetic is mine, not Norton’s, and the company would be doing readers a favor by publishing the base sizes so nobody has to estimate them.
One more thing the full report shows that the US release leaves out. American parents are more relaxed about AI fakes than parents elsewhere, at 67 percent worried against a 75 percent global average. American kids report cyberbullying at 25 percent against a 14 percent global average. Less worry, more incidents. That inversion is the actual story, and it did not make the press release.
What To Do This Week
For parents, three things beat another lecture about strangers:
- Ask your child what game chats they are in, by name, and who talks to them there
- Teach the escalation rule instead of the stranger rule: anyone who asks you to move to another app is the problem, friend or not
- Say out loud, once, that no image can get them in enough trouble to be worth hiding from you
That last one is the whole ballgame in a sextortion case. These extortions run on shame and a countdown clock. A child who believes they can come to you loses the extortionist their only leverage.
For everyone else reading this, the same synthetic media aimed at kids is aimed at your brand. Fake testimonials, cloned executives, and fabricated endorsement videos come off the same tooling. A whole category of startups now exists to fight deepfakes of real people, which tells you where this is headed. If you have not audited what a fabricated video of your CEO would cost you, that is a live exposure sitting on the same curve as this one.
Norton’s study is worth reading, and The Bots & The Bees is a reasonable resource for a conversation most parents are dreading. The research is also marketing for Norton 360, and that is fine. Vendors fund the survey work nobody else pays for.
Just read it the way you would read any document written by an interested party. Check the base sizes. Follow the links to the primary sources. And notice which findings made the summary and which ones stayed in the report.