Federal prosecutors are no longer waiting for Congress to write a bespoke statute for AI-generated child sexual abuse material. They are charging it now, under laws Congress passed more than two decades ago, and the first federal trial verdict on a purely synthetic image is poised to define the constitutional ceiling for every prosecution that follows.
The charging theory is not what most headlines suggest. Indictments are not built on 18 U.S.C. § 2256 alone. They pair section 2256's definition of child pornography, incorporated through the trafficking statute at 18 U.S.C. § 2252A, with separate obscenity counts under 18 U.S.C. § 1466A. That belt-and-suspenders approach exists because of a 2002 Supreme Court decision that still controls the field.
The statutory frame: two doors into the same courtroom
Section 2256(8)(B) defines child pornography to include images that are "indistinguishable from" an actual minor. Section 2256(8)(C) reaches morphed images that splice a real child's face onto sexual content. Section 1466A criminalizes obscene visual representations of the sexual abuse of children, including drawings and computer-generated images, and does not require that any real minor exist.
Both the "indistinguishable" prong and the 1466A obscenity statute trace to the PROTECT Act of 2003. Congress passed that law specifically to rebuild federal child pornography enforcement after the Supreme Court struck down the prior, broader regime.
Ashcroft v. Free Speech Coalition still controls
In Ashcroft v. Free Speech Coalition, 535 U.S. 234 (2002), the Court struck down the Child Pornography Prevention Act's ban on images that "appear to be" minors. The holding was direct: virtual child pornography that involves no real child is protected speech unless it qualifies as obscenity. The Court rejected the government's argument that such images could be banned because they might be used to groom victims or whet the appetite of offenders.
That ruling did not collapse the federal child pornography regime. New York v. Ferber, 458 U.S. 747 (1982), the foundational case carving child pornography out of First Amendment protection, rests on the harm done to real children during production. Ashcroft simply confirmed that the Ferber rationale does not stretch to images where no child was harmed in production.
Section 2256(8)(B)'s "indistinguishable from" language is Congress's narrower replacement for the invalidated "appears to be" standard. Section 1466A is the parallel track: it does not depend on Ferber at all, because obscenity has always been unprotected under Miller v. California.
The lead case: United States v. Anderegg
The most prominent test case is United States v. Anderegg, charged in the Western District of Wisconsin on May 20, 2024. The defendant is alleged to have used Stable Diffusion to generate sexually explicit images of minors and to have distributed them. The indictment charges counts under 2252A and 1466A in tandem, the same hybrid posture DOJ has used in subsequent announcements through 2024 and 2025.
That stacked structure matters. If a defendant defeats the 2252A counts by convincing a court that diffusion outputs are not "indistinguishable from" an actual minor, the 1466A obscenity counts still stand. If a defendant attacks 1466A on First Amendment grounds, the 2252A counts still stand. Prosecutors are deliberately giving themselves two independent routes to conviction.
Constitutional choke points the defense will press
Three constitutional arguments are forming the spine of the defense bar's response.
Overbreadth under Ashcroft. The defense will argue that 2256(8)(B), as applied to diffusion-model output, sweeps in protected speech because the model does not depict any identifiable child. Ashcroft's holding that speech cannot be banned because of secondary effects like grooming gives the argument its doctrinal core.
Void-for-vagueness on "indistinguishable." The defense will challenge whether a jury or an examiner can reliably distinguish a synthetic image from a photograph of a real minor, especially when diffusion models interpolate features from training data. The Stanford Internet Observatory's finding of real CSAM in the LAION-5B training set cuts both ways: it supports vagueness arguments (the model is not generating from nothing) while also supporting prosecution arguments (outputs may be derived from real victims).
Stanley v. Georgia private possession. Under Stanley v. Georgia, 394 U.S. 557 (1969), private possession of obscenity in the home is protected. The defense will argue that Osborne v. Ohio, 495 U.S. 103 (1990), which carved CSAM out of Stanley, did so because of harm to real children, and that the rationale evaporates for purely synthetic content. Most observers expect Osborne to foreclose the argument, but the question has never been squarely litigated in the AI context.
The circuit landscape
The closest existing appellate authority is United States v. Whorley, 550 F.3d 326 (4th Cir. 2008), in which the Fourth Circuit upheld 1466A obscenity convictions for Japanese manga-style cartoons depicting minors. Whorley gives the Fourth Circuit a clear doctrinal hook for synthetic imagery. The Ninth Circuit has handled comparable obscenity questions at the district level without producing a controlling appellate ruling on AI output.
A circuit split is plausible but has not yet matured. The first circuit-level ruling, whether out of the Fourth, Ninth, or Seventh, will likely shape DOJ's venue selection for years.
Prosecution strategy
Federal prosecutors are doing three things in parallel. First, stacking 1466A counts to neutralize Ashcroft challenges. Second, emphasizing training-data provenance, using the Stanford Internet Observatory work to argue that diffusion outputs are not "purely" virtual because the model was trained on real exploitation material. Third, pursuing plea agreements aggressively to avoid creating bad appellate precedent before the legal terrain is settled.
Plea math is driven by the U.S. Sentencing Guidelines at §§ 2G2.1 (production) and 2G2.2 (trafficking and possession). The base offense levels, enhancements for minors under 12, sadistic content, distribution, and number of images push exposure into double digits in years quickly, which gives prosecutors leverage to extract pleas that bypass appellate review.
What the TAKE IT DOWN Act did and did not do
The TAKE IT DOWN Act, signed May 19, 2025, criminalizes non-consensual intimate imagery, including AI deepfakes, and imposes a 48-hour platform takedown duty. It covers minors, but it is a civil and takedown overlay rather than a replacement for 2256 or 1466A. The criminal charging theory for fully synthetic CSAM still runs through the PROTECT Act framework.
What to watch
The next twelve to eighteen months will produce the pretrial motions that frame this area of law. The Anderegg docket and the parallel indictments announced by DOJ's Child Exploitation and Obscenity Section will generate the first written rulings on whether "indistinguishable from" survives an as-applied challenge to diffusion outputs, whether 1466A's obscenity standard reaches AI imagery, and whether private possession claims under Stanley have any traction. The Congressional Research Service's overview remains the cleanest map of how Ferber, Ashcroft, and Osborne fit together for anyone trying to follow the rulings as they come down.
For defendants and families, the practical takeaways are concrete. Charges in this space are almost always federal, almost always indicted in tandem on multiple statutes, and almost always carry mandatory minimums under 2252A and 2251. The legal questions are unsettled, but the sentencing exposure is not.
Related reading
Sources
-
18 U.S.C. § 2256 (definitions, including the "indistinguishable" prong)
-
Stanford Internet Observatory, Identifying and Eliminating CSAM in Generative ML Training Data
Note: This article contains AI-assisted content and has been reviewed by our editorial team.
IMPORTANT NOTICE: GavelDaily.com is an advertising and information service, not a law firm. Submitting this form does not create an attorney-client relationship. The information you provide will be shared with licensed criminal defense attorneys in your area who may contact you about your inquiry. All submissions are treated as confidential by our service, but attorney-client privilege does not attach until you formally retain an attorney. Any attorney you connect with through this service is independently responsible for their legal advice and representation. Free consultations are subject to individual attorney availability. If you are facing an immediate legal emergency, contact your local public defender's office or call 911.