Seventh Circuit Upholds Dismissal of AI CSAM Possession Charge

Seventh Circuit Upholds Dismissal of AI CSAM Possession Charge

The Seventh Circuit on August 25, 2026 affirmed the dismissal of one federal count charging private in-home possession of AI-generated obscene images of children, holding it was bound by Stanley v. Georgia and Ashcroft v. Free Speech Coalition. United States v. Anderegg, No. 25-1354.

Information last verified on September 5, 2026.

Status: Decided. The U.S. Court of Appeals for the Seventh Circuit issued its opinion in United States v. Anderegg, No. 25-1354, on August 25, 2026, after argument on October 30, 2025. It resolved the government's interlocutory appeal from the U.S. District Court for the Western District of Wisconsin (No. 3:24-cr-0050, Chief Judge James D. Peterson) and affirmed. The decision reaches only Count 4, the possession count under 18 U.S.C. § 1466A(b)(1), (d)(4). The district court denied the motion to dismiss as to the production, distribution, and transfer counts, and those counts were not before the panel.

Jurisdiction scope: This covers a single published federal appellate decision, binding on federal district courts in Illinois, Indiana, and Wisconsin, and only as applied to the facts of this prosecution. It does not repeal or facially invalidate 18 U.S.C. § 1466A, it does not reach federal offenses involving images of real children, and it has no effect on any state criminal statute.

What Happened

The prosecution began with a report from a platform, not a search. According to the government's account recited in the opinion, Meta Platforms reported the online transmission of potential child sexual abuse material to the CyberTipline of the National Center for Missing and Exploited Children in October 2023, under the reporting duty in 18 U.S.C. § 2258A. An Instagram user had sent a direct message containing what appeared to be AI-generated images to an account belonging to a minor, and the message history showed the user explaining to that minor how he had created the images by entering text prompts into Stable Diffusion, a text-to-image generative model (slip op. 3 to 4).

Investigators linked the account to Steven Anderegg, whose resume showed a software engineer with a background in AI models, and executed a search warrant covering his home, his personal laptop, two cell phones, and other devices. The government alleges he installed Stable Diffusion locally, added components that gave the program the ability to render human genitalia, and used curated prompts plus negative prompts that excluded adults to generate hundreds of images of prepubescent minors (slip op. 4).

He was charged on four counts. The first three, which the opinion cites to 18 U.S.C. § 1466A(a)(1), (d)(1), covered knowingly producing and knowingly distributing visual depictions of minors engaged in sexually explicit conduct, and knowingly transferring such material to a minor under sixteen. Count 4 charged knowing possession of at least one obscene visual depiction, under 18 U.S.C. § 1466A(b)(1), (d)(4) (slip op. 4 to 5).

Chief Judge James D. Peterson granted the motion to dismiss as to Count 4 and denied it as to the rest, holding § 1466A(b)(1) unconstitutional as applied because Stanley protects possession of obscenity in the home and because the government's justifications ran into Free Speech Coalition. The government took an interlocutory appeal of that single dismissal (slip op. 2 to 3, 5).

Two concessions shaped the appeal. The first appears in a footnote:

"The government concedes that the images for which Anderegg is charged do not depict an actual child, nor can they be linked to an actual child." United States v. Anderegg, No. 25-1354 (7th Cir. Aug. 25, 2026), slip op. at 4 n.2 (citing Oral Argument at 21:43)

The second came at argument, where the government acknowledged that on Count 4 it planned to rely solely on the fact that Anderegg possessed the material in his home (slip op. 17, citing Oral Argument at 6:46). That answer disposed of the government's leading argument for distinguishing Stanley, which was that the interstate-commerce element in § 1466A(d)(4) made the offense something more than mere private possession. The panel treated that element as more jurisdictional than substantive, noting in a footnote that jurisdictional elements "normally have nothing to do with the wrongfulness of the defendant's conduct" and declining to rest a constitutional distinction on one (slip op. 17 n.5). The panel also rejected the government's reading of Stanley as limited to obscene material involving adults, observing that Stanley turned on the location, the home, and not on the contents of the material (slip op. 18).

That left the question the panel called narrow: does the First Amendment allow the government to proscribe the in-home possession of obscene virtual CSAM (slip op. 17). Because the charged images depict no actual child, the panel concluded that New York v. Ferber and Osborne v. Ohio do not directly apply to this case, and then worked through each justification the government offered. Grooming risk, normalization of abuse, shrinking the market for real material, and the growing difficulty of proving an image depicts a real child were, the panel concluded, the same arguments the Supreme Court considered and rejected in Free Speech Coalition (slip op. 20 to 23).

The panel did not pretend to like the result:

"Given the ever-accelerating march of imaging-generation technology, we have some misgivings about applying Free Speech Coalition here, but 'unless we wish anarchy to prevail within the federal judicial system,' we are dutybound to follow it." United States v. Anderegg, No. 25-1354 (7th Cir. Aug. 25, 2026), slip op. at 23 (quoting Hutto v. Davis, 454 U.S. 370, 375 (1982))

The disposition follows from that premise. Because Anderegg is accused of possessing obscene material at home, his circumstances fall within Stanley; because every argument for removing the material from Stanley's reach had been rejected in Free Speech Coalition, § 1466A(b)(1) is unconstitutional as applied to him. Judgment affirmed (slip op. 23 to 24).

Judge Lee, joined by Judge Kolar, wrote separately to say the Supreme Court should revisit the area if a suitable case arises, and to identify a theory the government did not raise in this appeal. Note the limiting sentence that follows it, which is doing as much work as the observation itself:

"Thus, unlike the virtual CSAM at issue in Free Speech Coalition, the market for AI-generated virtual CSAM images may impact the demand for actual CSAM on which the AI models can train. ... In this case, the record is insufficient to evaluate these complex issues, and, of course, we are bound by the dictates of Free Speech Coalition." United States v. Anderegg, No. 25-1354 (7th Cir. Aug. 25, 2026) (Lee, J., concurring), slip op. at 27

Judge Pryor did not join that concurrence.

What the Law Actually Says

18 U.S.C. § 1466A is titled "Obscene visual representations of the sexual abuse of children." Congress added it through section 504(a) of the PROTECT Act, Pub. L. No. 108-21, on April 30, 2003, 117 Stat. 680, after Free Speech Coalition invalidated the Child Pornography Prevention Act's "appears to be" definition. Its structure matters to reading this case correctly.

Subsection (a) reaches anyone who, in a circumstance described in subsection (d), knowingly produces, distributes, receives, or possesses with intent to distribute a visual depiction of any kind, including a drawing, cartoon, sculpture, or painting. Subsection (b), headed "Additional Offenses," reaches knowing possession of the same material. Each subsection then splits into two prongs. Prong (1) covers a depiction that "(A) depicts a minor engaging in sexually explicit conduct; and (B) is obscene." Prong (2) covers a depiction that "is, or appears to be, of a minor engaging in graphic bestiality, sadistic or masochistic abuse, or sexual intercourse" and that "lacks serious literary, artistic, political, or scientific value." Penalties for subsection (a) offenses run through 18 U.S.C. § 2252A(b)(1); subsection (b) offenses run through § 2252A(b)(2).

Only § 1466A(b)(1), the possession-of-obscene-material prong, was before the Seventh Circuit. Prong (b)(2) was not at issue, and neither was any charge under subsection (a).

Three more pieces of the section are worth knowing. Subsection (c) states that it is "not a required element of any offense under this section that the minor depicted actually exist," which is why the statute reaches purely synthetic images at all. Subsection (d) lists five jurisdictional circumstances; the possession count used (d)(4), which asks whether the depiction was mailed, shipped, or transported in interstate or foreign commerce, or produced using materials that were. Subsection (e) supplies an affirmative defense that is available only to a subsection (b) charge, and only where the defendant possessed fewer than three such depictions and, promptly and in good faith and without retaining or letting anyone other than a law enforcement agency access them, either took reasonable steps to destroy each one or reported the matter to a law enforcement agency and afforded it access to each one. Subsection (f)(1) defines "visual depiction" to include any "digital image or picture, computer image or picture, or computer generated image or picture."

The two Supreme Court decisions the panel called controlling are old and narrow.

Stanley v. Georgia, 394 U.S. 557, decided April 7, 1969, reversed a Georgia conviction for knowing possession of obscene film reels found in the defendant's home. The Court accepted that states "retain broad power to regulate obscenity" but held "that power simply does not extend to mere possession by the individual in the privacy of his own home." Its most quoted sentence is the reason it keeps resurfacing: "If the First Amendment means anything, it means that a State has no business telling a man, sitting alone in his own house, what books he may read or what films he may watch."

Ashcroft v. Free Speech Coalition, 535 U.S. 234, decided April 16, 2002, addressed the CPPA's ban on material that "is, or appears to be," a minor in sexually explicit conduct, including computer-generated images. The Court distinguished Ferber on the ground that the child-protection rationale attaches to the production process: in contrast to material that is itself the record of abuse, "the CPPA prohibits speech that records no crime and creates no victims by its production." It also rejected, one by one, the grooming, whetting-the-appetite, market-elimination, and prosecutorial-difficulty rationales, while leaving open that a "significantly stronger, more direct connection" between virtual material and child exploitation might change the analysis.

Anderegg sat between those cases. As the panel framed it, Ferber and Osborne rest on harm to real children, Free Speech Coalition removes that rationale where no real child is depicted, and Stanley then governs what remains: obscenity possessed at home.

The state-law layer sits entirely outside this decision, and for most readers it is the more consequential layer. Wisconsin, where this prosecution is pending, has its own body of synthetic-media law, which we track on our page covering Wisconsin's deepfake statutes and, more broadly, in our survey of how the state regulates artificial intelligence. Legislatures in other states have moved on synthetic sexual imagery on their own timelines; our state-by-state deepfake law tracker is the place to compare them. On the adjacent federal question of nonconsensual intimate imagery, see our coverage of the TAKE IT DOWN Act and of state nonconsensual intimate imagery statutes.

Analysis: Why This Matters

The following is analysis from the Recording Law Editorial Team.

The most common way to misread this decision is to treat it as a statement about AI-generated abuse imagery. It is not. It is a statement about what a court of appeals is allowed to do when a 1969 privacy holding and a 2002 virtual-imagery holding both point one way and the facts of the case in front of it fall squarely inside both. The panel said so twice, once at the top of the opinion, where it wrote that it had "some concerns about the lines these cases draw, but we are not free to redraw them ourselves" (slip op. 3), and again at page 23 in the passage quoted above. Those sentences are the panel's assessment of the precedent it was applying, not an endorsement of the result that precedent produced.

The scope limits are doing enormous work here, and each of them is a variable that could change the outcome in the next case. The ruling is as-applied, which means the panel examined only these facts and expressly declined to consider "any set of hypothetical facts under which the statute might be unconstitutional" (slip op. 5). The government conceded that no actual child is depicted or identifiable, which took Ferber and Osborne off the board. The government also conceded it would rely solely on home possession, which took the commerce theory off the board. Change any one of those and the case is a different case. A prosecution involving morphed images of a real child, for example, sits outside this reasoning entirely; Free Speech Coalition itself flagged the morphing provision as one it was not considering because such images "implicate the interests of real children."

The concurrence is the part of the opinion most likely to be cited in the next round of litigation, and it is worth being precise about what it does. It does not hold anything. It identifies a factual theory, that generation of synthetic material may increase demand for the real material used to train models, and then says the record cannot support evaluating it. That is a roadmap, not a ruling: Free Speech Coalition left the door open for "a significantly stronger, more direct connection," and the concurrence is pointing at where a party might try to build one. The panel also noted that its own review of the literature returned inconclusive results, citing United States Sentencing Commission studies (slip op. 21 to 22 n.6). Any future record on this point would have to be built with evidence, not assertion.

For a reader who came here asking whether AI-generated sexual imagery of children is lawful, the honest answer is that this decision resolves one federal possession charge in one circuit and answers almost nothing else. The production, distribution, and transfer counts in this very prosecution were left standing. Federal law covering material that depicts real children was not touched. And criminal liability under state law is a separate question with fifty separate answers, which is exactly why the state-law tracking pages linked above tend to matter more to an ordinary reader than the federal doctrinal fight does.

How This Affects You

This decision does not create a safe harbor. As of September 5, 2026, its practical reach is limited in ways that are easy to state:

It applies to federal prosecutions in Illinois, Indiana, and Wisconsin, and only to the specific charge it addressed. In other circuits it is persuasive authority that a district court may consider and is free to disagree with.

It does not affect the offenses in 18 U.S.C. § 2252A or other federal offenses involving visual depictions of actual children, and the government's concession that no actual child is depicted was a precondition of the analysis, not a general rule.

It does not affect charges for producing, distributing, or transferring the same material, including the three counts still charged in this prosecution.

It does not displace state criminal law. A state statute reaching synthetic sexual imagery of minors operates independently of § 1466A and was not before the court.

What Happens Next

The Seventh Circuit affirmed, so Count 4 remains dismissed and the case returns to the Western District of Wisconsin on the counts the district court left intact. The ordinary avenues after a published panel decision are a petition for panel rehearing or rehearing en banc and, after that, a petition for a writ of certiorari. As of September 5, 2026 we have not confirmed on a primary source that any such petition has been filed, and we do not treat the absence of a docket entry we have not personally read as evidence either way.

The concurrence names the two events that would actually move this doctrine. The first is a case with a developed evidentiary record on the connection between synthetic material and the demand for real material, which is the "significantly stronger, more direct connection" Free Speech Coalition said it would take. The second is Supreme Court guidance, which the concurrence asked for in terms. Neither has happened, and we will not speculate about how either would come out.

This is general legal information, not legal advice. It describes one published federal appellate decision from the U.S. Court of Appeals for the Seventh Circuit, covering federal prosecutions in Illinois, Indiana, and Wisconsin, and the federal statutes and Supreme Court precedents that decision applies. It is not a statement of the law of any state, and it does not address any individual situation. Details verified as of September 5, 2026. Consult a licensed attorney in the relevant jurisdiction for advice on specific facts.

Related articles

• Wisconsin's deepfake statutes
• How Wisconsin regulates artificial intelligence
• State-by-state deepfake law tracker
• The TAKE IT DOWN Act
• State nonconsensual intimate imagery statutes

Last updated: 2026-09-05. Details verified as of 2026-09-05.