On August 25, 2026, the U.S. Court of Appeals for the Seventh Circuit ruled that a Wisconsin man cannot be prosecuted for privately possessing AI-generated images depicting child sexual abuse, as long as the images stay in his home and do not depict a real, identifiable child.

Let's be precise about what happened, because headlines have blurred it. This was not the Supreme Court. It was the Seventh Circuit, one level below, and the judges said plainly they had no choice. They were bound by old Supreme Court precedent. Two of the three judges on the panel wrote a separate opinion asking the Supreme Court to step in and fix what technology has broken. That is not a court endorsing this material. That is a court telling us the law is broken and only a higher authority can repair it. Laws are choices we make as a country, and we can choose differently.

The case is United States v. Anderegg. Steven Anderegg used AI to generate hundreds of hyper-realistic images of prepubescent children in sexually explicit situations. He was charged with four crimes: producing the images, distributing them, transferring one to a minor under sixteen, and possessing them. A lower court had already dismissed the possession charge on First Amendment grounds while letting the other three stand. The Justice Department appealed. The Seventh Circuit sided with the lower court and upheld the dismissal of that single charge.

The other three charges remain. Producing, distributing, and sending this material to a child are still crimes. Only private, unshared possession was struck down. In practice, that means a person can generate as many of these images as he wants and keep them to himself with no legal exposure — the only thing separating him from prosecution is whether he ever hits "share."

Three Supreme Court cases tie the Seventh Circuit's hands. Stanley v. Georgia (1969) held that the government generally cannot punish someone for privately possessing obscene material in their own home — a ruling about adult privacy and adult material. New York v. Ferber (1982) and Osborne v. Ohio (1990) carved out an exception for CSAM involving real children, because real children are harmed in its creation and that harm continues every time it circulates. Ashcroft v. Free Speech Coalition (2002) then struck down a law banning "virtual" child pornography — computer-generated or actor-based images with no real child involved — ruling that because no real child was harmed, the Ferber exception didn't apply.

Congress tried to close that gap with the PROTECT Act of 2003, criminalizing obscene depictions of minors "whether or not the minor actually exists." The Seventh Circuit ruled that law cannot reach private, in-home possession, because doing so collides with Stanley and Free Speech Coalition. That is the full chain: obscenity is protected at home, unless real children are involved, and AI-generated children don't count as real. Congress's fix doesn't survive that chain either.

Judge John Lee wrote the ruling. He and Judge Joshua Kolar then wrote a separate concurring opinion specifically asking the Supreme Court to revisit this area of law — something appellate judges rarely do. Lee noted that nearly twenty-five years after Free Speech Coalition, AI-generated CSAM is now "virtually indistinguishable" from material created through the actual abuse of a real child. Justices O'Connor and Thomas predicted exactly this problem back in 2002, warning that advancing technology would eventually make it impossible to tell real abuse material from fake, and that this might justify restricting even "lawful" virtual depictions.

Lee wrote that this day may have already arrived. He cited data showing that AI-generated images reported to the National Center for Missing and Exploited Children's Cyber Tipline are straining law enforcement, because even fake images consume investigative resources and make it harder to find real victims. Most disturbing: he cited research showing a major AI training dataset contained actual CSAM. That means some "fictional" AI-generated abuse material may be built, in part, on the abuse of real children, laundered through a machine into something legally possessable.

Lee did not pretend this was a small concern. He called preventing child sexual exploitation a government objective "of surpassing importance." But he said the court's hands were tied. That authority belongs to the Supreme Court alone.

In 2002, "virtual" child pornography meant crude graphics or adult actors dressed to look younger — nothing anyone could mistake for real abuse. That assumption is gone. Today's AI tools generate images the judges themselves call indistinguishable from the real thing, built in some cases from datasets that contained real abuse material.

And normalization isn't a side effect — it's the mechanism. Telling people that generating and privately keeping sexualized images of children is legal, as long as they don't share it, doesn't eliminate the demand. It hands people a private, personal loophole and tells them the only risk left is getting caught sharing it, or getting pulled over on the way to a friend's house with it still on their phone. That is not protecting speech. That is teaching predators how to stay legal while feeding an appetite that does not stay contained.

No one making this argument is attacking adult sexual expression. Adults can dance nude. Adults can create and share sexually explicit material with other consenting adults. That is protected, and it should stay protected. The question here is different: does the sexualized depiction of a child — real or AI-generated — deserve that same protection? Our answer, until AI exposed this gap, has always been no. We drew that line because children cannot consent, and because normalizing any version of their sexualization, "fictional" or not, creates cover and demand for the real thing.

The First Amendment has never been unlimited. The Supreme Court has already recognized categories that fall outside it, including obscenity and CSAM involving real children. This ruling doesn't prove the First Amendment protects everything — it exposes a technical crack, built for 2002-era technology, that AI has slipped through. We are allowed to close that crack. Adults get to express themselves with other adults. Children — real or photo-realistically simulated — are not a category anyone gets to claim as protected expression.

No legal system built by human beings is flawless. The judges who wrote this ruling are not villains — their concurring opinion shows judges genuinely troubled by the outcome they felt bound to reach. They identified the danger in detail and then told us who can fix it: the Supreme Court. That is not a reason to accept the outcome. It's a roadmap. We should push for the case, or the legislation, that puts this question back in front of the Supreme Court — not to restrict adult expression, but to say what most Americans already believe: AI-generated sexual images of children are not protected speech, wherever they're kept.

AI image generation is getting cheaper, more accessible, and more realistic every year. As more people learn that private possession sits in a legal gray zone, more will test that boundary. Law enforcement, already stretched thin identifying real victims, will spend more time sorting AI fakes from real abuse material — the exact burden Judge Lee warned about in his own opinion. That is not a fringe worry. It is what two sitting federal judges put in writing days ago.

This article sits beside our reporting on the FBI's child exploitation enforcement numbers, and that pairing is intentional. The FBI is doing real work: arrests are up, cases opened are up, and multiple nationwide surge operations have rescued hundreds of children and put hundreds of predators behind bars over the past year. That's an agency using the tools it has, aggressively, in the direction we want.

This ruling shows the other half of the problem. You can staff up enforcement all you want, but if the courts hand predators a legal loophole to generate and privately keep this material, you are asking law enforcement to fight with one hand tied behind their back. The FBI can arrest every trafficker and every real-world predator it finds — and it should keep doing exactly that — but it cannot arrest its way around a constitutional ruling that says private possession of AI-generated child abuse imagery is protected speech. That fix has to come from the courts, or from Congress, not from federal agents working cases.

So the message is simple and it has two halves. Keep funding and supporting the agencies doing the work — the FBI's numbers prove that pressure gets results. And at the same time, close this loophole at its source, because no amount of arrests downstream fixes a law that legalizes the behavior upstream. Adults can dance nude. Adults can express themselves sexually with other consenting adults. But children — flesh and blood or generated pixel by pixel — are not part of that conversation. You leave children alone. Full stop. It's time for the Supreme Court to say so clearly enough that no other court ever has to write a concurring opinion begging them to fix it again.