A federal judge says AI-generated images of child sexual abuse are protected speech. He thinks the law should change.

On Aug. 25, 2026, Judge John Z. Lee of the U.S. Court of Appeals for the 7th Circuit did what binding Supreme Court precedent forced him to do: dismiss federal child sexual abuse material (CSAM) charges against a defendant who allegedly used Stable Diffusion, a publicly available text-to-image model, to generate thousands of photorealistic images depicting the sexual abuse of children. A decade ago, that conduct would have looked unambiguous. In United States v. Anderegg (No. 25-1354), Lee wrote that the panel's hands were tied by two 20th-century cases the Supreme Court has never overruled.

The unusual posture is part of the story. Lee was not presiding over a trial. He was sitting as a circuit judge reviewing the district court's denial of a pretrial motion to dismiss. Once the panel concluded that the indictment, on its face, charged conduct the First Amendment protects, the appeal had to succeed; the facts alleged by prosecutors were not in dispute at that stage, and the constitutional question was dispositive. The ruling is therefore less a judgment about a defendant than a doctrinal one about a statute.

The two precedents doing the work predate the technology by decades. Stanley v. Georgia (1969) held that the government cannot criminalize the private possession of obscene material in the home. Ashcroft v. Free Speech Coalition (2002) struck down parts of the 1996 Child Pornography Prevention Act that criminalized "virtual" child pornography, imagery not depicting an actual child, on the ground that the First Amendment does not let Congress ban speech simply because it looks like the protected categories of expression the statute was designed to reach. Read together, the cases draw a line: federal CSAM liability requires either a real victim whose abuse produced the image or conduct outside the home. Imagery produced by a generative model from text prompts, depicting no real child, falls outside that line, and the home-possession overlay means even keeping it is protected.

That is the rule Lee felt bound to apply. It is not the conclusion he endorsed. In the same opinion, he wrote that the court "now live[s] 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" and that "the lines these cases draw" may not survive that era. The 7th Circuit is, in effect, telling Congress and the Supreme Court that the doctrine is being outrun by the capability it was never written to test.

The ruling does not empty the field. Federal law still reaches the conduct in other directions: distribution to a minor (the indictment here alleged the defendant transmitted some images to a 15-year-old boy), production that crosses a jurisdictional line, and parallel state charges under pandering, obscenity, or distribution-to-minors theories that do not depend on the imagery depicting a real child. The Washington Post reported on Aug. 26 that prosecutors are likely to keep pursuing those adjacent lanes, and legal-press analysis from Reason's Volokh Conspiracy and the Wisconsin Law Journal reads the panel's warning as a green light for legislative redrafting rather than a clean constitutional exit.

The Justice Department's posture predates the ruling. In May 2024, after the defendant's arrest, Principal Deputy Assistant Attorney General Nicole M. Argentieri said that "using AI to produce sexually explicit depictions of children is illegal, and the Justice Department will not hesitate to hold accountability those who possess, produce, or distribute AI-generated child sexual abuse material." That statement described the Department's position at the time of arrest, not after the 7th Circuit's reversal; whether the Department will seek further review, and on what theory, is now open. A Snopes fact-check and the Gray News wire summary circulated in late August corroborate the procedural and doctrinal record; the CourtListener docket for No. 68541611 tracks any cert petition the government may file.

The next move is legislative, not judicial, in the near term. Anderegg is the first federal appellate test of pre-diffusion CSAM doctrine against photorealistic generative imagery. Without a cert petition the Supreme Court takes, or a new statute that defines "child pornography" to include virtual depictions and survives First Amendment review, the gap Lee flagged will keep widening as model fidelity improves.