On August 25, 2026, a three-judge panel of the Seventh Circuit ruled that the First Amendment protects the private, in-home possession of AI-generated child sexual abuse material, so long as the images do not depict an actual, identifiable child. The case is United States v. Anderegg, and the man at the center of it, Steven Anderegg of Holmen, Wisconsin, allegedly used the open-source tool Stable Diffusion to mass-produce hyper-realistic sexual images of prepubescent children. He was ultimately investigated after Meta flagged an Instagram account that had sent this material directly to a minor.
The court didn’t let him walk on everything. The panel’s ruling only reached the possession charge. The charges for producing, distributing, and transferring the material to a minor still stand. That distinction matters and I want to be honest about it before I get to why this ruling should still make your stomach turn.
The legal logic, stated plainly
The judges felt bound by Ashcroft v. Free Speech Coalition, a 2002 Supreme Court decision that struck down a federal ban on virtual child pornography. Back then, the government argued that virtual images were hard to distinguish from real abuse material and that banning both was the only workable way to protect actual children. The Court rejected that reasoning, essentially holding that speech doesn’t lose constitutional protection just because it resembles something illegal.
That was a defensible position in 2002, when “virtual” child sexual imagery meant crude digital composites or drawings, nothing that could function as a training tool for further abuse or be mistaken for a real victim by anyone with functioning eyes. Judge John Lee, writing for the panel, more or less admitted this gap out loud, noting that the image-generation technology of 2026 was likely unimaginable when the Court decided that case. He and Judge Kolar went further in a concurrence, flagging real concern about where the current legal lines sit and openly inviting the Supreme Court to redraw them. They were not celebrating this outcome. They were describing the box they were stuck in.
That’s the part that should unsettle you more than a straightforward bad ruling would. This wasn’t three judges who think AI CSAM is fine. This was three judges who looked at binding precedent, looked at what modern generative models can now produce, and concluded the law hasn’t caught up. The dissent Rehnquist wrote in 2002, warning that advancing technology would eventually make virtual and real abuse imagery indistinguishable, reads less like a historical footnote now and more like a forecast that just landed on schedule.
Why “no real child” was never the right question
The instinct behind the original Free Speech Coalition ruling was that the harm in CSAM comes from the abuse of an actual child during its creation. No victim, no crime, at least not one the government can constitutionally reach through obscenity law alone. That logic assumed the image itself was inert once no child was harmed to make it.
That assumption doesn’t hold anymore, and the Seventh Circuit panel gestured at exactly why. Court filings referenced research showing that some of the datasets used to train image-generating AI have contained actual CSAM. If that’s true even in part, the “virtual” image isn’t disconnected from real victims at all. It’s downstream of them. The model learned what abuse looks like from something that was never virtual in the first place. And separate from training data contamination, there’s a second harm nobody in this legal framework is pricing in: this material is being used to groom real children right now, sent to actual minors as a normalization tool, exactly as prosecutors say happened here. The image doesn’t need a birth certificate to do damage to a living child.
So when a court says the material is protected because no real child was depicted, it’s answering a question from 2002 with a 2002 fact pattern, while sitting on top of a 2026 problem that doesn’t resemble it anymore.
The gap between outrage and fix
It’s tempting to read this ruling as judicial cowardice or moral failure. I don’t think that’s quite right, and I think it’s worth resisting that read because it points you at the wrong target. Three judges said, on the record, that they think the law needs to change. What they can’t do is change it themselves. Circuit courts don’t get to overturn Supreme Court precedent because the precedent has aged badly. That’s not how the hierarchy works, and pretending otherwise doesn’t fix anything.
The actual fix is legislative and it’s already partly in motion. Forty-six states have passed laws specifically targeting AI-generated CSAM since this technology became a serious problem. At the federal level, the ENFORCE Act passed the Senate unanimously, aimed at closing exactly the gap this ruling exposed, and there’s a further push in Congress to eliminate the “no real child” defense outright. None of that helps Anderegg’s possession charge retroactively, and none of it stops a court from having to apply Free Speech Coalition until the Supreme Court either narrows it or Congress writes a statute built for the world AI created. But it tells you where the actual leverage is. Anger pointed at this one panel is anger spent on the wrong branch of government.
What this actually means
If you’re a parent reading this and thinking the legal system just told you your child’s safety takes a back seat to a defendant’s home privacy rights, you’re not wrong to feel that, but you’re aiming at the wrong culprit. This ruling is a symptom of a doctrine written for a world before generative AI existed, applied faithfully by judges who flagged their own discomfort with the outcome. The people who can actually fix this are in state legislatures and in Congress, and some of them are already moving. The rest of us have a job too, and it has nothing to do with waiting on the Supreme Court. Every layer of protection you build around a child, at home, on their devices, in how you talk to them about secrecy and grooming, matters more in a world where the legal system is still catching up to what technology can generate. The law will get here eventually. Kids don’t have the luxury of eventually.