The video your family sent you is right about one thing

My sister-in-law sent us a video last week.

Sixty seconds. A woman talking straight into the camera, clearly upset. A law was just passed, she says. AI sexual images of children are legal now. Anybody can take a photo of your child, turn it into one, post it, and there’s nothing you can do. Stop posting pictures of your kids.

She sent it to us because of this site. She figured it was something we should be telling people about. She was right about that, and that’s the only reason you’re reading this, so I want to say it before I take the video apart.

Snopes checked the underlying claim on August 29 and rated it True. A federal appeals court did rule that the First Amendment protects a man who privately possessed AI-generated child sexual abuse images. That happened. Anybody who tells you the whole thing was invented is wrong.

What Snopes also said, in its own words, is that “the claim was true, but it lacked crucial context.”

The context is most of the story.

Nothing changed in 2026

The Seventh Circuit didn’t invent a new rule. It applied two Supreme Court decisions, one from 1969 and one from 2002.

Stanley v. Georgia, 1969, says a state can’t make it a crime to privately keep obscene material inside your own home.

Ashcroft v. Free Speech Coalition, 2002, is the one that matters here. Congress passed the Child Pornography Prevention Act in 1996 to outlaw computer-generated sexual images of children. The Supreme Court struck that part of it down. Material that doesn’t depict a real child isn’t child pornography under the law, and it’s protected speech. The decision covered digital images on purpose, not by accident.

So the protection the video described has been sitting there for twenty-four years. It wasn’t passed in August. It wasn’t passed at all, by anyone, ever. A court struck down the attempt to ban it, back when the technology at issue was crude 3D rendering.

What the court actually decided

Steven Anderegg was arrested in 2024 on four counts. A federal district court in Wisconsin dismissed one of them. The government appealed just that one. On August 25, 2026, Judge John Z. Lee of the U.S. Court of Appeals for the Seventh Circuit affirmed the dismissal (United States v. Anderegg, No. 25-1354).

Count 4. Possession, under 18 U.S.C. section 1466A(b)(1).

Three counts are still alive and headed for trial in Wisconsin. Count 1, production. Count 2, distribution. Count 3, transfer to a minor under sixteen. The appeals court never touched their merits, because nobody asked it to.

One count, then. Possession, inside a home, of images depicting children who don’t exist.

The judge doesn’t like his own ruling

This is the part the viral clips left out, and it’s the most alarming thing in the opinion, precisely because it’s so measured. Page 3:

We now live in an age where GenAI models can render images depicting the abuse of virtual children that are virtually indistinguishable from those depicting the abuse of actual children. This case illuminates how this evolving technology complicates the lines drawn by the Supreme Court in Stanley, Osborne, and Free Speech Coalition. Indeed, in Free Speech Coalition, the Supreme Court addressed the scope of First Amendment protections for virtual CSAM, but that was nearly twenty-five years ago, and the image-generation technology available today was likely unimaginable back then. Given the relentless advancement in artificial intelligence models, we have some concerns about the lines these cases draw, but we are not free to redraw them ourselves.

A federal judge saying the rule is wrong and he lacks the power to fix it, is a real warning. It’s a better one than the video gave, and it didn’t need a single exaggeration to get there.

Lee wrote separately, in a concurrence, that “recent data suggests that the influx of AI-generated images reported to the NCMEC CyberTipline imposes increasing burdens on law enforcement.” Worth one caution on that, since we hold ourselves to it here: a January 2026 Stanford study found that none of Amazon’s 380,000 generative-AI-flagged CyberTipline reports in the first half of 2025 actually involved AI-generated material. The strain on law enforcement is real. Counting it from raw report totals overstates it.

The Justice Department declined to say whether it will take the case to the Supreme Court.

The three claims in the video

A law was passed. No. A court decided a case, and decided it on precedent older than most of the parents sharing the clip.

Anybody can take a photo of your child and make one of these. This is the opposite of the ruling. Both the 2002 decision and this one turn on whether a real child is depicted. A photo of your child produces an image of a real, identifiable child, and that’s the category the court left completely alone.

They can post it and it’s legal. Posting is distribution. Distribution is Count 2, still pending, still going to trial.

Colorado, and the part worth getting exactly right

Colorado passed Senate Bill 25-288 in 2025, not 2026. Governor Polis signed it on June 2, 2025, and it took effect that August. It added AI-generated depictions to the criminal definition of sexually exploitative material at C.R.S. 18-6-403, and it reaches an “identifiable child,” meaning someone identifiable “by virtue of the individual’s face, likeness, or other distinguishing characteristic, including a unique birthmark or other recognizable feature.”

The first version would have banned all sexually explicit AI imagery of children whether or not a real child existed. The ACLU and the state public defender’s office opposed that scope, in testimony to Senate Judiciary in April 2025, and lawmakers narrowed the bill. That’s how the line ended up where it is.

Patrick Mueller, a spokesperson for the Colorado District Attorneys’ Council, explained the whole shape of it to Snopes in an email. Here it is in full, because the first two sentences are the ones that go missing when people quote this:

AI-generated Child Sexual Abuse Material is not outright illegal. If the AI generates CSAM that does not depict an identifiable child, it is not considered illegal under Colorado law. If the material is a mashup of multiple children, or real and unreal children, it would be illegal if any part of the image depicts a real, identifiable child. So, for example, an image that has a distinguishing birthmark from a real child on a leg but an AI-generated face would still be illegal since part of the image depicts an identifiable child. Again, identifiable does not mean prosecutors need to know the child’s name, just that the child is identifiable from the depiction that was created.

Read the third sentence twice. Nobody can launder a real child out of an image by regenerating the face. Any real, identifiable piece of your child makes the whole thing illegal in Colorado.

And that last line of his isn’t just his opinion, it’s in the statute. C.R.S. 18-6-403(9): “A prosecution brought pursuant to this section does not require proof of the actual identity of an identifiable child.” Prosecutors don’t have to know who your child is.

So what’s the actual gap

Wholly invented children. Nobody real, no trace of anyone real, kept privately. That’s it. That’s thin in Colorado and now thin in the Seventh Circuit, and a federal judge put in writing that it troubles him.

That’s the thing to be angry about, and it’s the thing to write your legislator about. It isn’t what the video described.

What the video got right

Be careful what you post. True before August 25, true after.

And something quieter that almost nobody picked up on. She said roughly twenty accounts are impersonating her, she reports them, and the platform finds no violation. That’s real, it’s happening to ordinary people right now, and no court ruling caused it.

The chatbot moment

She asked an AI whether it’s illegal for a non-parent to post photos of a child. It errored. She tried another device. It errored again. Her conclusion was that the AI is refusing to discuss it.

What almost certainly happened is that a question stacking “child,” “photos,” “posting,” and a sexual-abuse frame tripped a safety filter. Those filters are blunt, and they misfire constantly on exactly this kind of question. Ask a librarian the same thing and you get an answer. Ask a machine and you get a wall, and the wall means nothing.

A machine declining to answer is evidence that a machine declined to answer. That’s all it is. The instinct to read silence as a cover-up is going to cost somebody a lot more than one frightening evening.

Run it across the stone

The Touchstone Test asks four things. Look, trace, confirm, question the ask.

This one fails at trace, and it takes about four minutes.

Trace: a legal change has a source, and the source is either a bill with a number or a case with a court, a date, and a judge’s name. “The law was just passed” has none of those attached to it. Ask which law. If there’s no answer, you’re done.

Confirm: does the claim exist anywhere besides the video? It does. The Washington Post covered the ruling on August 26 under the headline “Federal judge warns law is being left behind by AI child sex abuse images,” which carries the judge’s reluctance right in the title. What it doesn’t carry, and what fell off everywhere downstream, is that three of the four counts survived.

The fourth question is worth naming even though this isn’t a scam. The woman in the video wasn’t asking for money and wasn’t selling anything. She read a headline, believed it, and warned other mothers. That isn’t fraud. It’s what accurate reporting turns into after four or five shares.

This isn’t even the first time on this subject. In July 2026, Snopes rated Mostly False a viral claim that seventeen Colorado Republicans voted against banning AI-generated child sexual images. The real House vote against was thirteen Republicans and nine Democrats, on a bill the Senate had passed unanimously, and the “seventeen” appears to have come from a different bill in a different year. Two distortions of the same topic in three months. This subject generates them, because it’s frightening and because the actual law is full of lines that don’t fit in a caption.

If it ever touches your family

Since May 19, 2026, covered platforms have had to remove a nonconsensual intimate image, including an AI-generated one, within 48 hours of a valid removal request. That’s the TAKE IT DOWN Act, Public Law 119-12. The request can come from the person in the image or from someone authorized to act for them, which covers a parent acting for a child.

Get this part right, because it’s the part people get wrong. You send the removal request to the platform, not to the government. Use the platform’s own reporting tool. If the platform blows the 48 hours, then you report the platform to the Federal Trade Commission at takeitdown.ftc.gov. The FTC enforces the deadline. It doesn’t take the image down for you.

You don’t need a lawyer for any of it. If a minor is involved, NCMEC’s CyberTipline is the other door, and the FTC’s own guidance points there.

Write that down somewhere you’ll find it later. Nobody researches a takedown process calmly on the day they need one.

And about the photos

Post fewer of them. Tighten the accounts. Do it because you decided to, not because a video scared you into it.

The next frightening video will show up with the same certainty in its voice, and it will be right about something too. Four minutes, four questions, before you send it to your family.