A federal appeals court ruled Tuesday that the First Amendment protects a person’s right to privately possess AI-generated child sexual abuse images, as long as the material does not depict a real child and stays inside the home.
The decision came from a three-judge panel of the U.S. Court of Appeals for the Seventh Circuit in a case called U.S. v. Anderegg. Judge John Z. Lee wrote the opinion, joined by Judges Doris Pryor and Joshua Kolar.
The case involves Steven Anderegg, 42, of Holmen, Wisconsin. The Justice Department charged him in 2024 with producing, distributing, and possessing obscene visual depictions of minors engaged in sexually explicit conduct, and with transferring obscene material to a minor under 16. Prosecutors say he used Stable Diffusion, a widely available AI image generator, to create more than 13,000 hyper-realistic images of prepubescent children in sexually explicit situations. Prosecutors have not alleged that any of the images depict a real, identifiable child.
Anderegg asked the court to dismiss the charges on First Amendment grounds. In February 2025, U.S. District Judge James D. Peterson in the Western District of Wisconsin let three charges proceed but threw out one: the charge for possessing the material in his home. Federal prosecutors appealed that dismissal. The Seventh Circuit has now upheld it.
What the appeal covered is the part getting lost as this circulates. The Seventh Circuit ruled only on the home possession charge. The charges against Anderegg for producing and distributing the material are still pending. So is the separate charge for allegedly sending similar AI-generated sexual images to a 15-year-old boy over Instagram. The court did not rule that making this material is legal, did not rule that sharing it is legal, and did not dismiss the case against him.
The legal reasoning runs through two Supreme Court decisions, both of which predate generative AI by decades. The first is Stanley v. Georgia in 1969, which held that a person has a constitutional right to possess obscene material in the privacy of their own home, even when the state can ban selling or distributing it. The second is Ashcroft v. Free Speech Coalition in 2002, which held that virtual child sexual abuse material, meaning depictions that do not involve any real child, is not child pornography under the law and cannot be banned on the same basis. Actual child sexual abuse material remains illegal to possess anywhere under Osborne v. Ohio.
Put those together, and the panel concluded that the statute Anderegg was charged under, Section 1466A, is unconstitutional as applied to his private, in-home possession of images that no child was harmed to create. The court noted that every argument the government made to pull this material out from under Stanley’s protection had already been rejected by the Supreme Court in Free Speech Coalition.
The scope is narrower than the headline suggests. Riana Pfefferkorn, a policy fellow at Stanford, pointed out that the ruling does not protect downloading images from a third-party website, and does not protect carrying a phone or laptop containing the images outside the house. That distinction has already been tested. In a January 2026 case, U.S. v. Yener, a court followed the Anderegg reasoning but reached a different result because the defendant’s phone containing the material was found on him at a diner and in a storage unit rather than at his residence. The court held that the protection turns on the privacy of the home specifically, not on private possession generally, and declined to treat a cell phone outside the home as the constitutional equivalent of a residence.
The judges who wrote the opinion are not comfortable with it. Lee wrote that given the relentless advancement in artificial intelligence models, the panel has concerns about the lines these cases draw, but that they are not free to redraw them. Lee and Kolar then filed a separate concurring opinion asking the Supreme Court to weigh in, writing that they would benefit from additional guidance on the intersection between the First Amendment and virtual child sexual abuse material if the right case comes along. They noted that nearly 25 years after Free Speech Coalition, that ruling still controls in a world where AI-generated material is virtually indistinguishable from material produced by physically abusing a child, an outcome several Justices predicted back in 2002.
That gap is what the case is actually about. The laws and the precedent were written for an era when creating a convincing fake required either a real child or a level of skill and expense that put it out of reach. That is no longer true, and the volume shows it. The National Center for Missing and Exploited Children received roughly 4,700 CyberTipline reports with a generative AI component in 2023. In 2024, that number hit 67,000. Between January and September 2025, according to Fallon McNulty, who runs the center’s exploited children division, the tipline received more than a million. Pfefferkorn has cautioned that the generative AI checkbox on the reporting form can mean several different things, so those figures should be read as a trend rather than a precise count of images. Either way, the direction is not in dispute.
McNulty has said the nearly indistinguishable nature of the content makes victim identification extremely difficult, which is the practical harm even when no child was photographed. Investigators cannot tell at a glance whether an image is synthetic or evidence of a child being abused right now, and every hour spent sorting that out is an hour not spent finding a real child.
The FBI’s position has not changed. Kiersten Schiliro, a senior technical advisor at the bureau, said earlier this year that even fully synthetic material with no child victim is still a crime and that the Justice Department will prosecute it. The Seventh Circuit ruling narrows one specific charge in one specific circumstance. It leaves the rest standing. Anderegg’s remaining charges are still in front of the court.
