In United States v. Anderegg, the 7th Circuit held a federal possession charge could not be brought for private, at-home possession of AI-generated sexual images of children. It left the production, distribution, and transfer charges in place. Law enforcement agencies and investigators must plan now for when (not if) they are called upon to handle cases involving AI-generated (or potentially AI-generated) child sexual abuse material (CSAM).
Image-generation software powered by artificial intelligence can now produce photorealistic sexual images of children without a camera and, in some cases, without involving any real children. That’s an uncomfortable problem, but it’s the unfortunate reality of policing during the age of AI.
Practically every law enforcement agency has had cases dealing with child sexual abuse material, whether through a direct investigation or a discovery on a device seized for some other reason. AI-generated material is now part of that same caseload.
This raises two critical questions:
- Is AI-created child sexual content illegal? Mostly, yes, although there are some important caveats. The legal situation is now less certain than it was a year ago.
- How does an investigator know whether an image or video depicts a real child? The short answer is that investigators can no longer rely on their own eyes and instincts regarding what’s real and what isn’t.
This discussion will cover what is happening in Anderegg — what courts have done so far, why every agency should expect to encounter this material, and actions officers and investigators should take when they do.
The Reality of Synthetic Sexual Material Involving Children
In the fields of law enforcement, criminal justice, and child advocacy, the term “child sexual abuse material” (CSAM) was adopted to replace “child pornography” to make an important point: Each example of this type of content constitutes evidence of child sexual abuse.
The rise of AI-generated material that is virtually indistinguishable from “genuine” CSAM raises an obvious question. If no real child was abused to make it, can it even be considered CSAM? Answering this question gets even harder when you acknowledge that AI systems are “trained” using images and video — sexual and non-sexual — of real children. Because of this, a case can be made that the material victimizes every one of the children whose images helped the AI produce the offending material. Just as a child can’t consent to sexual activity, children also can’t consent to having their images used to train AI image-generation systems.
It’s objectively true that practically every new technology — including the printing press, photography, motion pictures, home video, the internet, and now AI — is inevitably used to create and disseminate sexual content. As a result, legislators in many states have passed laws targeting AI-generated or computer-edited CSAM.
Enough Abuse, a child advocacy organization, counts 47 states with such laws. [1] Whether those laws are effective is a different question that almost no one has studied. The only analysis we know of is a 2025 Stanford report that reviewed state laws and interviewed people who deal with this very modern problem. According to the researchers, many of these laws did not anticipate cases where the offender is a child (such as teens using “nudify” apps on their peers), and legislative staff said prosecutors sometimes had to get creative or decline cases before their state passed a law. Most of the laws were too new to have even been enforced when the report was written.
Of course, AI image- and video-generating capabilities have continued to improve since this report was released, and the volume of material has grown with it. The unhappy truth is that laws, case law, and regulation are slow to catch up.
Overview of the Anderegg Case
United States v. Anderegg, 189 F.4th 576 (7th Cir., 2026)
In October 2023, an Instagram user sent direct messages containing apparent CSAM to the account of a minor. Meta flagged the images and reported the account to the National Center for Missing & Exploited Children (NCMEC), which generated a CyberTip. The messages also included text prompts explaining how the images had been made using Stable Diffusion, a widely available AI image generator.
Investigators from Wisconsin’s Division of Criminal Investigation identified Steven Anderegg of Holmen, Wisconsin, as the owner of the Instagram account. The agency executed a search warrant and found photorealistic images of prepubescent children. The government later conceded none of the images depicted, or could be linked to, a real child.
A federal grand jury charged Anderegg with three counts under 18 U.S.C. § 1466A, the federal statute on obscene visual representations of the sexual abuse of children, and one count (transfer to a minor) under 18 U.S.C. § 1470:
- Knowingly producing the images.
- Knowingly distributing the images.
- Transfer to a minor. Knowingly sending the material to a minor under 16.
- Knowingly possessing at least one such image.
Section 1466A applies whether or not the minor depicted actually exists, which is why prosecutors relied on it in this prosecution.
Anderegg moved to dismiss. On Feb. 13, 2025, Chief Judge James Peterson of the U.S. District Court for the Western District of Wisconsin dismissed only the possession count, holding the statute unconstitutional as applied to private possession at home of images that depict no real child. The court relied on Stanley v. Georgia (394 U.S. 557 (1969)), which protects private, in-home possession of obscene material. The other three counts went forward.
The government appealed the dismissal of the possession count, and on Aug. 25, 2026, the 7th Circuit affirmed the lower court’s dismissal. Writing for the panel, Judge John Z. Lee concluded that Stanley and Ashcroft v. Free Speech Coalition (535 U.S. 234 (2002)) applied. Ashcroft held that Congress could not ban virtual images made without a real child merely because they appear to depict minors. In a separate concurrence, Judges Lee and Joshua P. Kolar wrote that AI-generated imagery is now virtually indistinguishable from imagery of real abuse, pointed out that a major dataset used to train image generators contained hundreds of CSAM images, and said the court “would benefit from additional guidance from the [U.S.] Supreme Court.” Even so, the judges said they were bound by precedent.
This particular ruling is an as-applied holding about private, in-home possession of wholly synthetic obscene images. It does not touch the counts involving production, distribution, or transfer to a minor, which remain pending. It also does not apply to images of real children, including AI-altered images of identifiable minors. And it binds only courts in the 7th Circuit (Illinois, Indiana, and Wisconsin), although defense attorneys elsewhere will likely cite it.
What the Courts Have Done So Far
Much of the legal framework around sexual content is decades old. Stanley (cited above) protects private possession of obscene material at home. New York v. Ferber (458 U.S. 747 (1982)) and Osborne v. Ohio (495 U.S. 103 (1990)) apply to the possession of CSAM involving real children, even at home, because making it requires abusing a child. Ashcroft held that purely virtual material lacks that link to a real victim. Congress responded by narrowing the federal definitions in 18 U.S.C. § 2256(8) and by enacting § 1466A, which relies on the obscenity standard.
Note that, in the narrow August decision in the ongoing Anderegg case, the 7th Circuit panel held that § 1466A(b)(1) was “unconstitutional as applied to Anderegg because, under Stanley v. Georgia, individuals have a First Amendment right to possess obscenity in the privacy of their home.”
Other courts have declined to extend the circuit court’s Anderegg ruling beyond the limits of the case. For example, in United States v. Yener (S.D. Fla. Jan. 21, 2026), the court denied a motion to dismiss a § 1466A(b) possession count because the defendant kept his images on phones located outside his home (i.e., at a diner and a storage unit). Stanley, the court reasoned, depends on the privacy of the home. Where the files were found can matter.
For cases in which real children are involved, prosecutions continue. In May 2026, a federal court in the Southern District of Ohio sentenced Austin Pittman to 14 years in prison for using AI to alter images and videos of real children “to whom Pittman had access.” Investigators identified the real victims through a search warrant on Pittman’s phone. In June 2026, a federal jury found Jeremy Koppenhaver guilty of one count of receipt of CSAM and one count of production of obscene visual representations of the sexual abuse of children. The Justice Department said “[a]ll the images of child pornography depicted real children,” adding that some of the images were created using AI tools.
The Supreme Court has not ruled on this topic, and we can find no other relevant federal or state appellate decision as of the publication date. Public reports indicate the three remaining counts against Steven Anderegg have not been resolved.
In Congress, the Senate passed the Enhancing Necessary Federal Offenses Regarding Child Exploitation (ENFORCE) Act in December 2025, which would increase criminal penalties for CSAM to include offenses involving AI-generated or modified depictions of minors, while also removing certain statutes of limitations. The House has not yet brought the bill up for a floor vote, with House leadership keeping it “held at the desk.”
As mentioned above, approximately 47 state legislatures have passed laws criminalizing AI-generated or computer-edited CSAM. The language of the laws varies widely. For example, Louisiana’s “unlawful deepfakes” law concerns only images created to look like a real, identifiable person. Arkansas’s law does not require an identifiable child. Laws in Texas (S.B. 1621 and H.B. 2700) and Alabama use a “virtually indistinguishable” standard, and Pennsylvania covers material that “appears to authentically depict” a child.
Whether state possession statutes can withstand a Stanley challenge is still an open question. Officers and agencies should be aware of exactly what their own state’s statute covers.
It Is a Matter of When, Not If
For law enforcement, the question is not whether an agency will encounter AI-generated CSAM, but when. That’s because the numbers are heading in just one direction.
NCMEC’s CyberTipline received about 4,700 reports involving material created with generative AI in 2023. That number rose to 67,000 in 2024 and more than 400,000 in 2025. Many of those tips reach agencies without a clear answer on whether a real child is involved.
The material may show up in several ways. A platform may file a CyberTip, as Meta did in the Anderegg case. A forensic examination of a phone or computer seized in an unrelated case may turn it up. A school may report students using “nudify” apps on classmates, which can mean both the offender and the victims are minors. A family may report sextortion built on a fake explicit image of their child.
Tools that agencies currently rely on can be inadequate when it comes to synthetic CSAM. Hash matching helps identify only known images. A newly generated image will not have been identified before, and a model running locally on someone’s computer leaves no trail with a platform. As a result, investigators are more likely to find this material directly on devices than through a CyberTip.
Nor can officers assume that synthetic means victimless. NCMEC received more than 12,000 reports in 2025 involving CSAM found in AI training data, and it says deepfakes can start from an ordinary photo of a real child. A real child’s likeness or a real child’s abuse may be behind an image that looks computer-made. Until that is ruled out, the safe assumption is that a victim may exist.
Then there is the workload. One offender can generate hundreds of images, and each one has to be examined by investigators. Time spent on synthetic images competes with time spent finding real victims.
What to Do When You Find It
The difficult truth is that it’s impossible to settle the question of whether a piece of content is real or synthetic just by looking at it. The concurrence in Anderegg says as much. Because of this, the correct response for law enforcement must be to protect the evidence, the case, and possible victims — all at the same time.
The steps below are just a starting point. Your agency counsel, your local Internet Crimes Against Children (ICAC) task force, and your jurisdiction’s prosecutor should confirm them against state law and local practice.
Every path requires documentation. The only “branch” in the process is determining whether a real child could be involved.
- Stop, secure, and limit viewing. Officers should not forward, copy, screenshot, text, or upload the material, and should not run it through public AI or image-detection websites. Officers or investigators should notify a supervisor, and the forensic unit’s guidance should govern how the device is seized.
- Treat the material as potentially real until shown otherwise. A wrong guess in either direction has a cost: a missed victim, a wasted search, a compromised case.
- Confirm legal authority before opening anything flagged. As my co-author and I wrote in March, courts currently disagree on whether files automatically flagged by a provider can be viewed by police without a warrant. Unless counsel confirms the circuit allows it, officers should obtain a warrant. In addition to the images themselves, the warrant should cover image-generation software, model files, prompt histories, chat messages, and linked accounts.
- Preserve and document. Officers should record where the files were found, on what device, and who had access, and keep the chain of custody intact. Generation evidence matters because how an image was made can matter as much as what the image depicts. After Yener and Anderegg, whether the files were kept in a home can determine whether a charge can be sustained.
- Bring in the specialists. The ICAC task force or the agency’s digital forensics unit can authenticate the images, run hash checks, and use NCMEC’s resources to see whether the images match known victims.
- Work the victim question. If the photo or likeness of a real child may have been used, treat the case as a child safety case first. That means identifying the child, notifying the child or the child’s parents or guardians, requesting that platforms take the images down, and coordinating with the school if the child is a student. If the person who made or shared the images is also under 18, follow agency policy and state law on juvenile suspects, and involve the prosecutor early.
- Talk to your prosecutor early. Whether the image depicts a real child, an altered image of a real child, or a wholly synthetic image affects the charges that can be brought. State law, 18 U.S.C. § 2252A, § 1466A, and the circuit’s view of Anderegg all matter.
- Ask targeted interview questions. How was the material made, with what tools and prompts, from what source images, and were photos of real children used? The answers, and any claim that the material is synthetic, should all be recorded.
- Document every check, result, and decision. The record should show which tool was used, which version, what it showed, and what remained uncertain.
- Protect the investigators. Limit exposure to this content to those who have a functional need to view it. Provide wellness support to anyone in your agency who is required to review this material.
The legal authority and the victim question should come before any debate about whether an image or video clip is real.
Tools and Resources
No officer has to make these decisions alone. The contacts and tools below are worth identifying before a case arrives.
- NCMEC’s CyberTipline. Suspected AI-generated exploitation can be reported there, and NCMEC makes reports available to the appropriate law enforcement agency. NCMEC has also labeled more than 158,000 images and videos as AI-generated, which helps with triage.
- Take It Down. This NCMEC service lets a victim or family anonymously request removal of explicit images of a minor, real or AI-generated, from participating platforms.
- NCMEC’s Child Victim Identification Program. Investigators can use it to compare images against those of known, identified child victims.
- ICAC Task Force Program. This national network of state and local task forces offers training, technical help, and investigative support on cases involving CSAM and synthetic CSAM.
- Digital forensics tools. Products such as Amped Authenticate and Magnet Verify analyze metadata, compression, and sensor noise to help judge whether an image came from a camera or another source. Court acceptance of AI-based detectors is still limited, so examiners should expect to explain their methods and their limits. The Department of Homeland Security has also begun testing an AI-image detector from Hive AI, but it has not yet been accepted as a standardized tool.
- Your prosecutor and state attorney general’s office. They can advise on charging under state law, § 2252A, and § 1466A, and on how the circuit has treated Anderegg.
- Enough Abuse’s state law tracker. It summarizes whether, and how, each state covers AI-generated CSAM.
Have a Plan Before the Case Arrives
For most agencies, the best way to start is by naming an ICAC contact and a prosecutor for these cases, confirming what the state statute covers and updating definitions, warrant templates, and evidence-handling procedures to include AI-generated material. After that, it’s critical to train detectives, patrol officers, and school resource officers on what to do in the first 10 minutes after material like this is uncovered.
Courts may take years to settle the key question of whether AI-generated sexual imagery is a “victimless crime.” The officer or investigator who finds this kind of material does not have that luxury. Law enforcement personnel should act immediately, treat the material as real until it is proven otherwise, and be able to explain every step that followed.
Notes and Resources
- According to Enough Abuse, the only U.S. jurisdictions that haven’t passed laws against artificially generated or computer-edited CSAM are Massachusetts, Ohio, Vermont, and Washington D.C.
- Blog Articles
