The Scariest Number in Child Safety Was Wrong

Intro

Welcome to The Tech Radar for Parents. I'm your host, Weston, brought to you by Nicholas D. West Tech LLC. This is your briefing for the week of August seventeenth, 2026.

Quick note before we start. Today we talk about sextortion, and about the statistics on abuse imagery. I'm not going to describe how any of it works. And if you or your child are struggling, the 988 Suicide and Crisis Lifeline is free, confidential, and open right now. Call or text 9 8 8.

Now. Fourth week in a row, I went looking for a new viral challenge. There isn't one. Fourth week in a row, I went looking for new slang. There isn't any.

But on Tuesday, in a federal courthouse in Oakland, California, a lawyer stood up in front of eight jurors and said that Meta built Instagram and Facebook to keep children on them, quote, to the point of physical and mental harm.

And I want to tell you about a number. You have probably seen it. Four hundred thousand reports of A I generated child abuse imagery. It is the scariest number in this entire field. And this week I can tell you, with a primary source, that it does not mean what you were told it means.

Let's go.

The Good: The Scariest Number Was Wrong

Let's start with the good, because this week the good news is that something you were afraid of is smaller than you were told.

If you follow child safety at all, you have seen this headline: the National Center for Missing and Exploited Children received more than four hundred thousand reports last year tied to artificial intelligence. It got repeated everywhere.

Here is what a researcher at Stanford found when she went and looked.

Her name is Riana Pfefferkorn. In a letter dated January of this year, she examined the reports where the Generative A I box was checked. In the first half of 2025 there were about four hundred and eighty five thousand of them. And at least seventy eight percent involved no A I generated material at all.

Not less than you'd think. None.

Here's why. Roughly three hundred and eighty thousand of them came from one company. Amazon. Amazon was scanning the data used to train A I models, and inside that training data they found known abuse imagery of real children. Photographs that already existed, already matched by fingerprint. And they reported every single one. Correctly. That is exactly what they should have done.

So the A I connection was that the pictures were sitting inside the training data. Not that A I made them. That checkbox is a flag on a form. It is not a measurement.

Now, I want to be careful here, because it would be easy to hear the wrong thing. This does not mean A I generated abuse imagery isn't real. The Center's own number for offenders possessing or generating that material is more than one hundred and eighty two thousand. That is a real number and it is a horrifying one.

But one hundred and eighty two thousand is not four hundred thousand. And a detection system finding old crimes buried in a training set is not the same thing as a new crime wave.

The good news is twofold. The panic number was an artifact, and the reason we know is that scanning worked, a company reported what it found, and a researcher checked the math in public.

So when you see four hundred thousand in a school newsletter this fall, and you will, you now know something the person sharing it doesn't.

The Bad: Four States, Eight Jurors, and a Judge Who Decides

Now the bad. And this is the biggest child safety trial in the country.

On Tuesday, August eighteenth, opening statements began in Oakland, California, in the United States District Court for the Northern District of California. The judge is Yvonne Gonzalez Rogers.

Let me give you the numbers carefully, because almost every headline gets at least one of them wrong.

Twenty nine states sued Meta back in 2023 as a coalition. But four states are actually trying this case. California, Colorado, Kentucky, and New Jersey. Not twenty nine lawyers in a courtroom. Four states, going first.

And here is the part that matters most, and that I have not seen a single mainstream headline explain properly.

Judge Gonzalez Rogers convened an advisory jury. Eight people. An advisory jury does not decide this case. The verdict and the penalties ultimately rest with the judge.

Write that down, because in about two months you are going to see a headline that says jury finds against Meta, or jury clears Meta. Whichever way it goes, that headline is describing a recommendation. Not a ruling.

So what are the states arguing? Their counsel summed up Meta's business model in four words. Hook, hold, harvest, hide. Hook the kids, hold their attention, harvest their data, hide what the company knew. The core allegation is that Meta designed Facebook and Instagram to keep children on the platforms longer and longer, quote, to the point of physical and mental harm.

And I want to be clear about a word there. Allegation. This is what the states argued in an opening statement. Meta disputes it. An opening statement is not evidence, and nothing has been proven.

The first witness the states called is Arturo Bay-har, a former Facebook engineering director who became a whistleblower. Over the weekend before trial, the judge rejected Meta's attempt to keep him off the stand.

This runs six to eight weeks. Mark Zuckerberg is expected to testify. So is Adam Mosseri, who runs Instagram. Which means for the next two months you're going to be reading internal Meta documents in the newspaper. Read them as what they are. Evidence being tested in public, one side at a time.

The Ugly: Nine Hundred Forty-Two Million Dollars That Nobody Is Paying Yet

The ugly this week is about a landmark ruling, and about how little a landmark ruling means until the appeals are done.

You may have seen that Meta was ordered to pay five hundred and sixty seven million dollars in New Mexico. That figure is real, but it's only part of it.

The total is nine hundred and forty two million dollars, and it breaks into two pieces. Back in March, a jury imposed three hundred and seventy five million in civil penalties. That came from seventy five thousand willful violations of New Mexico's consumer protection law, at the five thousand dollar statutory maximum, each. Then on August sixth, the court added five hundred and sixty seven million for abatement. That's money to clean up the harm. Judge Bryan Biedscheid entered the final judgment.

And the order isn't only money. For five years, it would require Meta to stop sending overnight push notifications to users under eighteen, between ten P M and seven A M. Mandatory time limits for minors. Public like counts hidden by default for kids. A Child Safety Monitor, appointed by the court and paid for by Meta. And public compliance reports, filed with the court twice a year.

If you're a parent, that list probably sounds like most of what you've been asking for.

Here is the ugly part. None of it is in effect. Not one item on that list.

On Tuesday, the same day opening statements began in Oakland, Meta went into state court in Santa Fe and asked Judge Biedscheid to freeze enforcement of the entire judgment while it appeals, offering to post a bond of one point eight billion dollars instead.

That motion is pending. The judge has not ruled. And as of Thursday, Meta had not actually filed its appeal yet. It has thirty days from August sixth to do it.

So a company can be found liable, be ordered to change how its products work for children, and then post a bond and keep operating exactly as before while the appeal runs. That is not a scandal. That is how the system is designed to work. But it is the reason I'm asking you not to change one thing about how you supervise your kid's phone on the strength of a court order that is currently frozen.

Do not wait for the ten P M notification ban. Set your own.

The Decoder Ring: Five Things Being Reported Wrong

Now the Decoder Ring. Normally this is where I translate five slang words your kid is using.

Fourth week in a row: there is no new slang. So we're doing this instead, and honestly I think it's more useful. Here are five things being reported wrong right now, including one I need to correct from the coverage this very week.

One. Seven sixty four is not slang. I keep seeing it on lists of codes parents should know. It is not a code and it is not a word. Seven sixty four is the name of a violent online criminal network. The F B I's Boston office put out an advisory in February. More than three hundred and fifty subjects under investigation nationwide, all fifty six field offices involved. If you're scanning your kid's texts for the digits, you're looking for the wrong thing.

Two. Six seven is not the 2026 Word of the Year. Dictionary dot com named it Word of the Year for twenty twenty five, on October twenty eighth of last year. There is no twenty twenty six Word of the Year yet. And that article going around about teachers dreading the twenty twenty six, twenty twenty seven school year? Published in December. The twenty six, twenty seven is the school year, not the date.

Three. Six seven is not a drug code, a sex act, or a self harm reference. I have now seen all three claims. Dictionary dot com's own entry says it is, quote, impossible to define, and meaningless. It appears on no D E A slang list. It came from a rap track and a hand gesture. Sometimes a nonsense word is just a nonsense word.

Four. An advisory jury is not a jury. We covered this, but it's the one that's going to bite people in October. Eight people are listening in Oakland. They give an opinion. Judge Gonzalez Rogers decides.

Five, and this one is mine to flag. This week a search engine told me the Texas Attorney General sued Roblox and Discord on August nineteenth. Big news, if true. It isn't. I pulled both press releases directly. The Roblox suit is from November of twenty twenty five. The Discord suit is from May of this year. The August dates were links in a sidebar of unrelated stories. Both lawsuits are real. They're just not from this week.

That is exactly the mistake that ends up in a parent Facebook group as, did you hear what happened this week. Check the dateline. Always check the dateline.

Your Weekly Action Plan, Steps 1 and 2

Alright. Your action plan.

Step one. Set the overnight rule yourself, tonight. A court in New Mexico decided that Meta should stop pinging kids between ten P M and seven A M. That order is frozen. You are not frozen. Whatever phone your kid has, it can silence notifications overnight. Every phone made in the last decade does this, and it takes about ninety seconds in the settings. The court thought this mattered enough to order it. Don't wait to find out whether the appeal succeeds.

Step two. Retire the four hundred thousand number, and say why. If you've shared it, you're in good company. It was in every outlet. But now you know that at least seventy eight percent of those flagged reports involved no A I generated material, and that most of them came from one company doing exactly the right thing. The number to use is one hundred and eighty two thousand, and it's still awful. Correcting a scary statistic in your parent group chat is not downplaying it. It's the opposite. Numbers that fall apart under scrutiny are what make people stop listening to the real ones.

Your Weekly Action Plan, Steps 3 and 4

Step three. Teach the dateline check. This is the single most useful media literacy skill you can hand a middle schooler in five minutes, and you can practice it on me. Pull up any scary story about kids and tech. Find the date. Then ask: is this from this week, or is it from last year and going around again? I disqualified three stories this week on that test alone, including one that was exactly one year old to the day. Your kid will be better at this than you within a month.

Step four. Have the sextortion conversation before you need it. The guidance from the National Center for Missing and Exploited Children is precise, so let me give it to you precisely. If this ever happens, your child should get help before deciding whether to pay or comply. Not never pay. Get help first. Because as the Center puts it, cooperating or paying rarely stops blackmail. Say the whole sentence out loud to your kid this week. If someone ever has a picture of you and is threatening you with it, you come to me first, and you will not be in trouble. The threat only works while they think they're alone with it.

Closing

So, to recap the week of August seventeenth. No new challenge, no new slang, fourth week running. Opening statements in Oakland, where four states are trying the case in front of an eight person advisory jury and a judge who makes the final call. A nine hundred and forty two million dollar judgment in New Mexico that Meta is trying to freeze with a one point eight billion dollar bond. And the scariest number in child safety turning out to be a checkbox.

One thing I owe you before I go. I check five categories every week. Challenges, slang, influencers, apps and regulation, and scams and safety. Last week the influencer research failed and I told you so. This week it ran, and it came back empty for these seven days. California's child influencer bill, S B twelve forty seven, is real and still moving, but its last action was August tenth and it has not passed. It is not law. If anyone tells you California just protected kid influencers, they are ahead of the facts.

And two things I did not check this week. Whether any state law took effect in these seven days beyond that California bill, and the kidfluencer laws in Illinois, Minnesota, Utah, and Washington. I'd rather tell you where I didn't look than let you assume I looked everywhere.

If you or your child are struggling, call or text 9 8 8. Free, confidential, twenty four hours.

I'm your host, Weston, with Nicholas D. West Tech LLC. Check the dateline, set the overnight rule, and I'll see you next week.