Did a Federal Appeals Court Really Rule That AI-Generated CSAM Is Legal to Possess?

The viral claim contains a real and significant court ruling, but “AI-generated CSAM is legal” is much too broad.

On August 25, 2026, the U.S. Court of Appeals for the Seventh Circuit held that a federal possession law, 18 U.S.C. § 1466A(b)(1), was unconstitutional as applied to one charge based solely on Steven Anderegg’s possession, inside his home, of obscene AI-generated images that the government conceded did not depict and could not be linked to an actual child.

The court did not strike down §1466A as a whole. It did not establish a general right to produce AI-generated sexual depictions of minors, distribute them, send them to other people, transport or receive them, or possess imagery involving an actual child. And the appeal did not eliminate three other charges against Anderegg for production, distribution and transfer of obscene material to someone under 16.

The distinction is the entire story:

The Seventh Circuit held that the federal government could not use this particular possession provision to punish this particular form of mere, private, in-home possession. It did not hold that AI-generated CSAM is lawful in every context.

What exactly did the Seventh Circuit rule?

United States v. Anderegg arose from a four-count federal indictment in the Western District of Wisconsin. Prosecutors alleged that Anderegg used Stable Diffusion to produce synthetic imagery and charged him with four separate offenses. Those remain allegations unless and until proven.

The indictment confirms those four separate statutory theories.

That wording matters. It would be inaccurate to say the Seventh Circuit “upheld” Counts 1 through 3. Those counts were not the issue the appeals court was deciding. The district judge had refused to dismiss them, while the federal government’s interlocutory appeal challenged the dismissal of Count 4. The Seventh Circuit affirmed that dismissal.

The appellate panel itself emphasized how limited the question was:

“The question before us is narrow.”

The government also conceded at oral argument that, for Count 4, it planned to rely solely on the fact that Anderegg possessed the material in his home. That concession became critical to the constitutional analysis.

Why “AI child porn is legal” gets the ruling wrong

There are at least five separate qualifications hidden by that formulation.

First, this was an as-applied constitutional ruling, not a decision invalidating §1466A everywhere.

Second, it concerned mere possession, not production, receipt, transportation or distribution.

Third, it concerned possession inside a private home.

Fourth, the government conceded that the charged images did not depict an actual child and could not be linked to an actual child.

Fifth, the case involved one federal statute. It did not decide the constitutionality or applicability of every other federal or state law that can reach synthetic or manipulated imagery.

So there is a substantial difference between:

“The government cannot constitutionally prosecute this defendant under §1466A(b)(1) for this particular in-home possession.”

and:

“AI-generated CSAM is legal.”

The first accurately describes the holding. The second does not.

The decisive fact: no actual child was involved in the charged images

The distinction between a record of abuse involving a real child and a purely virtual depiction lies at the center of the Supreme Court precedents controlling the case.

The Seventh Circuit did not independently decide that the images were synthetic after a factual trial. Rather, the government conceded for purposes of the case that the charged images did not depict an actual child and could not be linked to one.

That removed one of the government’s most powerful constitutional justifications for criminalizing possession: protecting an identifiable victim whose abuse was required to create the image and whose continuing circulation can perpetuate that harm.

It is why the result in Anderegg cannot simply be transferred to a conventional photograph or video created through sexual abuse of an actual child.

Why can possession be protected when obscenity itself normally is not?

This is the counterintuitive part of the case.

The Supreme Court has long held that obscenity generally falls outside normal First Amendment protection. But it has also created an unusually narrow constitutional protection for possessing obscene material in one’s own home.

Four Supreme Court cases explain almost the entire result.

Stanley v. Georgia: government generally cannot criminalize mere possession of obscenity in the home

In Stanley v. Georgia in 1969, police searching a man’s home discovered obscene films. The Supreme Court held that the government could regulate obscenity broadly but could not make mere private possession in one’s home a crime.

The Court was explicit that its decision concerned possession in the special privacy of the home rather than public distribution or commerce. (Justia Law)

That does not mean obscenity itself suddenly became protected speech in all settings.

United States v. Orito: Stanley does not create a right to transport or distribute obscenity

Four years later, United States v. Orito made the boundary clearer.

The Supreme Court upheld federal authority over transporting obscene material in interstate commerce and explained that the privacy zone recognized in Stanley does not follow the material outside the home. (Justia Law)

That distinction is essential to understanding Anderegg:

A constitutional protection against punishment for merely having something in a private home does not automatically create a constitutional right to acquire it, transport it, distribute it or sell it.

That may sound strange as a practical matter, but it is the doctrine the Seventh Circuit had to apply.

New York v. Ferber and Osborne v. Ohio: actual child victims change the equation

The Supreme Court created a very different rule for material produced through exploitation of actual children.

In New York v. Ferber, the Court held that sexual depictions produced through the use of real children could be prohibited without satisfying the ordinary Miller obscenity test. The government’s interest was not merely regulating offensive ideas; it was preventing the exploitation of children used to create the material and the market supporting that abuse. (Justia Law)

Then came Osborne v. Ohio in 1990. The Supreme Court held that Stanley did not prevent a state from banning even private possession in the home of material depicting actual children.

The difference was the victim. The Court emphasized both reducing the market for material created through child exploitation and the continuing harm caused by the existence and circulation of a permanent record of abuse. (Justia Law)

Ashcroft v. Free Speech Coalition: purely virtual depictions are constitutionally different

In 2002, Ashcroft v. Free Speech Coalition confronted a federal law that had expanded the definition of prohibited material to encompass sexual images that merely appeared to depict minors, including computer-generated depictions created without actual children.

The Supreme Court struck the challenged provisions as overbroad. Importantly, the law swept in virtual material that did not necessarily satisfy the constitutional definition of obscenity.

The Court also refused to treat the real-victim rationale from Ferber and Osborne as automatically applicable when no real child was used to create the image. (Justia Law)

That does not mean Free Speech Coalition immunized all virtual sexual imagery involving minors.

That distinction is particularly important because the present §1466A does something the law struck down in Free Speech Coalition did not: it can require the material to be obscene.

How those cases produced the Anderegg decision

Put the precedents together:

The Seventh Circuit concluded that Anderegg’s Count 4 landed at the intersection of Stanley and Free Speech Coalition.

Because the prosecution was relying only on possession in his home, Stanley applied. And because the government conceded that no actual child was depicted or linked to the charged material, the justifications that took actual-child imagery outside Stanley under Osborne could not simply be transplanted to this case.

The court therefore concluded that §1466A(b)(1) was unconstitutional as applied to Anderegg.

What does 18 U.S.C. §1466A actually prohibit?

This point is another reason the shorthand “AI CSAM was legalized” fails.

Section 1466A sits in Chapter 71 of Title 18, the federal obscenity chapter. Its title is “Obscene visual representations of the sexual abuse of children.”

The statute expressly reaches computer-generated imagery. Depending on the subsection and jurisdictional circumstances, it prohibits conduct including:

• production;
• distribution;
• receipt;
• possession with intent to distribute; and
• knowing possession.

And Congress expressly provided that it is not necessary for the depicted minor actually to exist. The statute’s definition of a visual depiction also includes a computer-generated image or picture.

The Seventh Circuit did not erase those provisions.

It decided that §1466A(b)(1)’s possession prohibition could not constitutionally be applied to the specific form of in-home possession before it.

Even the district court refused to extend Stanley from possession to production

This is perhaps the clearest answer to claims that the ruling created a general First Amendment right to make such material.

Anderegg had also argued that Stanley should protect private production. The district judge rejected that argument.

Judge James Peterson reasoned that Stanley was a narrow holding about possession and that the Supreme Court had never extended it to a general constitutional right to produce obscenity in one’s home. He therefore denied the motion to dismiss Count 1, the production charge. (Wikisource)

The Seventh Circuit’s later decision did not reverse that portion of the case.

So even within Anderegg itself, there is a concrete judicial distinction between:

possessing the material privately at home

and

producing it.

What about receiving or downloading the material?

Anderegg does not establish a general right to receive or download obscene virtual imagery.

That is another place where headlines can turn a narrow possession rule into something much broader.

The Supreme Court’s post-Stanley cases have repeatedly distinguished mere possession inside a home from importing, receiving, transporting and distributing obscene material. Orito is one example. And the Seventh Circuit itself noted that the cases narrowing Stanley generally involved movement or distribution rather than in-home possession “in and of itself.”

The unusual posture of Count 4 mattered because prosecutors told the Seventh Circuit that they intended to rely solely on the possession in the home, rather than prove a distinct act of obtaining or transporting the charged material.

That means a future case involving downloading, receiving or exchanging material could present materially different facts.

But isn’t computer-generated imagery already included in the federal definition of “child pornography”?

Federal terminology here is more complicated than either side of the online debate usually acknowledges.

Section 2256(8), which supplies definitions for a different federal chapter, defines “child pornography” in several ways. It encompasses material produced through use of an actual minor, but it also includes certain digital or computer-generated imagery that is “indistinguishable from” an actual minor, as well as imagery created, adapted or modified to make an identifiable real minor appear to engage in sexually explicit conduct.

At the same time, §2252A contains an affirmative defense to several of its offenses when the alleged material was not produced using an actual minor, with an exception for the identifiable-minor category in §2256(8)(C).

The result is that statements such as:

“AI images cannot be child pornography under federal law”

are too categorical.

But the opposite formulation—

“Federal law treats every synthetic image exactly like a photograph produced through the abuse of a real child”

—is also wrong.

Most importantly, Count 4 in Anderegg was brought under §1466A, and that is the statute the Seventh Circuit’s holding addresses.

Was Anderegg the first appeals court to recognize this distinction?

No.

That is an important correction to some of the novelty surrounding the case.

In United States v. Ostrander in 2024, the Eleventh Circuit expressly summarized the existing Supreme Court doctrine this way: private possession in one’s own home of obscene material depicting virtual minors can receive constitutional protection when no real children are victimized.

But Ostrander did not win his challenge.

His case involved a broader facial challenge to §1466A, and the Eleventh Circuit concluded that the statute has many constitutionally valid applications—including transportation and possession outside a private home. His circumstances therefore did not produce the same result as Anderegg’s narrowly framed as-applied challenge.

So the best description is not:

“Anderegg invented a new constitutional right to possess virtual CSAM.”

Instead:

Anderegg appears to be the first regional federal appellate decision squarely applying the Stanley/Free Speech Coalition principle to an as-applied challenge involving generative-AI imagery where the prosecution’s possession theory was confined to the defendant’s private home.

That is still an important development.

Other cases show why the location and conduct matter

Several earlier appellate decisions reached different outcomes without necessarily contradicting Anderegg.

United States v. Whorley

The Fourth Circuit upheld convictions involving receipt of obscene anime-style cartoons under §1466A. It distinguished Stanley because the prosecution concerned movement of obscene material through channels of commerce rather than merely finding material sitting inside someone’s home. (Fourth Circuit Court)

United States v. Bowersox

The U.S. Court of Appeals for the Armed Forces upheld a §1466A possession conviction involving computer-generated imagery. But the defendant possessed it in a shared military barracks room, which the court declined to treat as the private home protected by Stanley. (United States Courts – ARM)

United States v. Buie

The Eighth Circuit upheld §1466A(b)(1) against facial overbreadth and vagueness challenges involving hand-drawn imagery. But Buie did not squarely decide the particular as-applied Stanley question that controlled Anderegg. (Leagle)

That distinction is why it is premature to describe Anderegg as creating a clean conflict among the federal circuits.

What about other generative-AI cases?

There has already been disagreement among federal trial judges over how Stanley applies when the material or the equipment used to make it has an interstate-commerce connection.

In United States v. Yeasley, a federal district judge in Idaho rejected a constitutional challenge involving generative-AI imagery, reasoning in part that §1466A’s interstate-commerce requirements materially distinguished the case from Stanley. (GovInfo)

A federal district court in Ohio took a similar position in United States v. Prater, rejecting the reasoning of the Wisconsin district court and emphasizing the interstate-commerce component of §1466A. (GovInfo)

But those are district-court decisions, not contrary rulings from the Ninth or Sixth Circuit.

And the Seventh Circuit directly confronted the commerce argument in Anderegg. Because the government had limited Count 4 to possession in Anderegg’s home, the appellate panel concluded that the statute’s interstate-commerce jurisdictional element did not transform the substantive conduct being punished into transportation or distribution.

So there is genuine lower-court disagreement around the edges, but not yet a clean regional-circuit split on identical generative-AI facts.

What if an AI image is based on an identifiable real child?

That is materially different from the facts the Seventh Circuit decided.

Federal law specifically recognizes imagery that has been created, adapted or modified to make an identifiable real minor appear to engage in sexually explicit conduct. Section 2256(8)(C) covers that category, and §2256 separately defines what qualifies as an identifiable minor.

The Supreme Court itself distinguished manipulated or “morphed” imagery involving actual children when discussing virtual imagery in Free Speech Coalition.

So the Anderegg holding should not be translated into:

“AI deepfakes of real children are constitutionally protected at home.”

The Seventh Circuit did not decide that question.

What if the AI model was trained on images of actual child abuse?

The Seventh Circuit’s concurrence raises exactly that issue—but does not answer it.

Judge John Lee, joined by Judge Joshua Kolar, wrote separately to argue that modern generative AI may have complicated assumptions underlying the Supreme Court’s 2002 decision in Free Speech Coalition.

The concurrence cited research indicating that a major image-generation training dataset had contained hundreds of CSAM images. It suggested that if markets for AI-generated material were shown to create demand for actual abuse material used in training, some of Osborne‘s real-victim and market rationales might become relevant again.

But the judges then made an essential qualification:

The record in Anderegg was insufficient to evaluate those questions.

That means the concurrence did not find that:

• Anderegg’s particular model was trained on actual CSAM;
• his charged images were derived from identifiable abuse images;
• generative-AI imagery necessarily increases demand for actual CSAM; or
• such a connection would automatically change the constitutional result.

Those are open questions identified by two judges, not holdings of the court.

Why did the judges rule this way if they were worried about the result?

Because a federal appeals court cannot simply discard binding Supreme Court precedent when new technology makes that precedent uncomfortable.

The panel acknowledged that present-day generative AI can produce imagery far more realistic than the computer imagery contemplated when Free Speech Coalition was decided in 2002.

The judges wrote that they had concerns about how the old doctrinal lines operate in the era of modern image generation—but that they were “not free to redraw them” themselves. They concluded that Stanley and Free Speech Coalition controlled the case and left any reconsideration of those precedents to the Supreme Court.

The separate concurrence was even more explicit in asking for future Supreme Court guidance.

That makes Anderegg less a declaration that Congress can never regulate AI-generated material and more a demonstration of what happens when 1969 and 2002 constitutional doctrine collides with 2026 technology.

Does the ruling apply nationwide?

Not as binding circuit precedent.

The Seventh Circuit is a regional federal appeals court covering federal district courts in Illinois, Indiana and Wisconsin. (Seventh Circuit Court)

Its published decisions bind federal district courts within that circuit on materially similar federal questions unless superseded by higher authority or later circuit proceedings.

But the decision:

• does not bind federal courts in the other regional circuits;
• does not repeal §1466A;
• does not itself invalidate state statutes;
• does not determine whether materially different AI imagery falls under another federal law; and
• could ultimately be affected by rehearing or Supreme Court review.

It would therefore be misleading to summarize the geographic effect as:

“AI-generated CSAM is now legal in Illinois, Indiana and Wisconsin.”

The lawfulness of particular conduct depends on what the person did, what the imagery actually is, what statute is invoked, where the conduct occurred and which jurisdiction’s law applies.

What the Anderegg ruling does—and does not—protect

The clearest way to understand the decision is as a scope map:

So did a federal appeals court rule that AI-generated CSAM is legal to possess?

In the narrow setting actually before the Seventh Circuit, yes: the court held that the federal government could not constitutionally apply §1466A(b)(1) to punish the mere, private, in-home possession charged in Count 4, where the government conceded that the AI-generated images did not depict and could not be linked to an actual child.

But that is not equivalent to saying “AI-generated CSAM is legal.”

The Seventh Circuit did not create a nationwide immunity for synthetic imagery. It did not protect production, distribution or transfer. It did not extend the ruling to imagery involving actual children or identifiable-child deepfakes. It did not invalidate §1466A as a whole. And it did not decide what happens when a future prosecution can establish a meaningful connection between generative-AI outputs and material created through actual child abuse.

The remarkable part of Anderegg is not that judges suddenly declared synthetic CSAM acceptable. It is that existing Supreme Court doctrine draws a constitutional line between a record created through the abuse of an actual child and purely virtual obscenity—and, for mere possession inside a private home, that distinction changed the outcome of a federal criminal charge.

Whether that line survives generative AI in its current form is now a question the Seventh Circuit has effectively placed before the Supreme Court.

References and Further Reading

Primary case records and statutes

United States v. Anderegg — U.S. Court of Appeals for the Seventh Circuit, Aug. 25, 2026 The controlling appellate opinion. It contains the majority’s narrow as-applied holding and the Lee-Kolar concurrence discussing unresolved generative-AI questions.

United States v. Anderegg — Federal Indictment, Western District of Wisconsin The primary charging document. It establishes the exact four counts and statutory provisions rather than relying on shorthand descriptions in subsequent coverage.

United States v. Anderegg — District Court Opinion and Order, Feb. 13, 2025 The ruling that originally dismissed Count 4. This source is a transcription of the filed district-court order and is particularly useful for the court’s refusal to extend Stanley to the production charge.

18 U.S.C. §1466A — Obscene Visual Representations of the Sexual Abuse of Children The statute directly at issue. It expressly includes computer-generated imagery and states that an actual minor need not exist.

18 U.S.C. §1470 — Transfer of Obscene Material to Minors The statute underlying Count 3, which was not affected by the Seventh Circuit’s possession ruling.

18 U.S.C. §2256 — Federal Definitions for Chapter 110 Important for terminology. It distinguishes actual-minor imagery, certain computer-generated imagery that is indistinguishable from actual minors, and modified depictions involving identifiable minors.

18 U.S.C. §2252A — Certain Activities Relating to Material Constituting or Containing Child Pornography Shows why the broader federal statutory picture is more complicated than §1466A alone, including the affirmative defense applicable to some material produced without actual minors.

Controlling Supreme Court precedents

Stanley v. Georgia, 394 U.S. 557 (1969) Established the narrow constitutional protection against criminalizing mere private possession of obscenity inside one’s home.

United States v. Orito, 413 U.S. 139 (1973) Clarified that Stanley does not create a corresponding constitutional right to transport obscenity in interstate commerce.

New York v. Ferber, 458 U.S. 747 (1982) Explained why material produced through exploitation of actual children can be prohibited independently of the ordinary obscenity rules.

Osborne v. Ohio, 495 U.S. 103 (1990) Permitted criminalization of in-home possession of actual-child pornography, distinguishing it from the ordinary obscenity protected by Stanley because actual victims are involved.

Ashcroft v. Free Speech Coalition, 535 U.S. 234 (2002) Struck down overbroad federal provisions reaching virtual depictions produced without actual children and rejected several government rationales later revisited in Anderegg.

Related appellate and generative-AI cases

United States v. Ostrander — Eleventh Circuit, 2024 Important because the Eleventh Circuit expressly recognized the constitutional distinction for purely virtual material privately possessed at home, while rejecting Ostrander’s broader facial challenge.

United States v. Whorley — Fourth Circuit, 2008 Shows the importance of receipt and movement through commerce rather than mere home possession.

United States v. Bowersox — Court of Appeals for the Armed Forces, 2013 Upheld a §1466A conviction involving virtual imagery because a shared military barracks room was not treated as a Stanley-protected private home.

United States v. Buie — Eighth Circuit, 2019 Upheld §1466A against facial overbreadth and vagueness challenges involving drawn imagery; it did not squarely resolve the same narrow as-applied home-possession theory as Anderegg.

United States v. Yeasley — District of Idaho, 2025 A generative-AI district-court case taking a different view of §1466A’s interstate-commerce component before the Seventh Circuit decided Anderegg.

United States v. Prater — Southern District of Ohio, 2025 Another district-level generative-AI decision emphasizing interstate commerce. It helps explain the developing disagreement but does not itself create a circuit split.

Editorial currency note: Anderegg was decided only on August 25, 2026. The opinion is currently published Seventh Circuit precedent, but the litigation is recent enough that rehearing proceedings, Supreme Court review or subsequent developments in the underlying criminal case could change its posture. Federal and state laws governing synthetic and manipulated sexual imagery are also developing rapidly. This article should be reviewed if the Seventh Circuit issues a further order, the government seeks Supreme Court review, or Congress materially amends the relevant statutes. (Seventh Circuit Court of Appeals)