Seventh Circuit Gives Narrow Protection to Private Possession of AI-Generated Child Images

The Seventh Circuit upheld one dismissed possession charge on August 25, despite prosecutors pursuing three other charges involving AI-generated abuse images. Across Google News, that distinction can disappear behind alarming summaries suggesting the court broadly legalized harmful synthetic content.

It did not.

The court decided a narrow constitutional question in United States v. Anderegg. The ruling concerns private, in-home possession of obscene images that depict no identifiable real child. It does not create a general right to produce, distribute, transmit, sell, or receive such material.

That boundary matters because the defendant still faces allegations involving production, distribution, and sending material to a minor. Those accusations involve actions beyond private possession. The appellate court left those charges intact.

The decision nevertheless exposes a serious conflict. Supreme Court cases from 1969 and 2002 protect certain private possession and virtual expression. Modern image generators can produce material that investigators struggle to distinguish from records of real abuse.

Families therefore face two risks at once. Synthetic abuse can cause real harm when it targets an identifiable child. Meanwhile, misleading coverage can obscure which conduct the court protected and which conduct remains prosecutable.

What the Seventh Circuit Actually Decided

The court invalidated one statute only as applied to one possession charge, not as a universal rule for AI-generated sexual material.

Steven Anderegg was charged in the Western District of Wisconsin under several provisions of federal law. Prosecutors alleged that he used Stable Diffusion, a text-to-image generator, to create synthetic sexual images depicting children.

The government conceded that the images covered by the charges did not depict an actual child. Prosecutors also acknowledged that investigators could not connect those images to an identifiable child.

According to the appellate opinion, Anderegg used written prompts and negative prompts when generating the material. A negative prompt tells an image model what elements to exclude from its output.

Meta reportedly alerted authorities after material was sent through Instagram to a minor. A later search uncovered hundreds of related synthetic images on the defendant’s devices.

The indictment included four categories of alleged conduct:

• Producing visual depictions of minors engaged in sexually explicit conduct

• Distributing those depictions

• Transferring such material to a person under 16

• Possessing at least one obscene depiction covered by 18 U.S.C. Section 1466A(b)(1)

The district court dismissed only the fourth count. It concluded that applying the possession statute to private, in-home possession violated the First Amendment under existing Supreme Court precedent.

The government appealed that dismissal. A three-judge Seventh Circuit panel affirmed it after reviewing the constitutional issue from the beginning, without deferring to the district court’s legal conclusion.

This was an as-applied ruling. That term means the court evaluated how a statute operated against this defendant under the alleged facts. The panel did not erase the statute or declare every possible application unconstitutional.

The distinction also limits the ruling geographically. The Seventh Circuit establishes federal appellate precedent in Illinois, Indiana, and Wisconsin. Other federal appeals courts can reach different conclusions until the Supreme Court resolves a conflict.

Even within the Seventh Circuit, the decision protects only the conduct addressed by the court. The panel repeatedly separated possession inside the home from production, distribution, transportation, and transfer.

The court also accepted the indictment’s allegations as true for this stage of the case. It did not conduct a trial or decide whether Anderegg committed the remaining alleged offenses.

Local Wisconsin coverage reported that Anderegg was arrested in 2024 and was 42 when the article appeared. It also emphasized that three charges remain pending.

That procedural posture is essential. A ruling about one count is not an acquittal on the entire indictment. Nor is it a broad endorsement of the underlying material.

Why the Google News Headline Is Easy to Misread

A short Google News summary can turn a precise constitutional boundary into the false impression that courts legalized an entire category of abuse.

The phrase “protected by the First Amendment” sounds sweeping without its surrounding conditions. Here, those conditions do most of the legal work.

The ruling applies to possession, not every interaction with the images. It addresses material kept inside a home, not content posted publicly or shared through a platform. It also concerns images that do not depict or identify an actual child.

Remove any of those conditions, and the legal analysis changes.

This is why the headline should not be read as permission to create or circulate synthetic child sexual abuse material. The court explicitly noted that possessing obscene material at home does not establish a related right to buy, receive, transport, or distribute it.

The decision rests mainly on two Supreme Court cases. The first is Stanley v. Georgia, decided in 1969. Stanley recognized a narrow right to possess obscene material privately in one’s home.

The second is Ashcroft v. Free Speech Coalition, decided in 2002. That case rejected parts of a federal law targeting virtual depictions that did not involve real children.

The Supreme Court distinguished virtual material from child sexual abuse material recording an actual child. Images of real abuse preserve the victimization and can compound harm through continued circulation.

Congress has broader authority to prohibit material created through the exploitation of real children. The government can also criminalize conduct surrounding obscenity, including many forms of distribution and commerce.

In Anderegg, prosecutors argued that today’s synthetic images create new enforcement problems. Photorealistic output can make investigators search for children who do not exist. Suspects might also claim that images of real victims are synthetic.

Those concerns are substantial, but the panel found them legally familiar. The Supreme Court considered similar arguments in 2002, including predictions that virtual and real images would become indistinguishable.

The Seventh Circuit said it remained bound by that decision. Lower federal courts must follow controlling Supreme Court precedent, even when later technology weakens its original factual assumptions.

Judge John Z. Lee wrote separately, joined by Judge Michael Kolar, to request further Supreme Court guidance. Their concurrence did not change the result. It explained why the governing framework now deserves reconsideration.

This creates an unusual message for readers finding the story through Google News. The judges upheld the defendant’s constitutional argument while openly questioning whether older precedent fits current technology.

That is a conflict between judicial authority and technological reality. It is not a declaration that the judges considered the conduct harmless.

Private Possession Is Not Production or Distribution

The central legal contest is private possession versus outward conduct, not child protection versus unrestricted AI use.

Stanley protects a narrow zone inside the home. Later Supreme Court decisions have repeatedly refused to convert that protection into a right to obtain, transport, or distribute obscene material.

That boundary explains why the remaining allegations matter. Prosecutors say Anderegg did more than store generated files privately. They accuse him of producing material, distributing it, and transferring it to a 15-year-old boy.

The Seventh Circuit did not dismiss those counts. Their continued existence shows why “AI images are legal” is an inaccurate description of the decision.

Generation itself can involve actions distinct from passive possession. A person chooses prompts, adjusts settings, evaluates outputs, and saves selected files. The process can therefore support a production allegation under a statute that reaches virtual depictions.

Distribution creates another legal category. Uploading a file, sending it through a messaging service, or transferring it to another person takes the conduct outside the home’s protected sphere.

Sending sexual material to a minor adds a direct human interaction. That conduct no longer depends only on whether a depicted child exists. A real recipient can experience grooming, coercion, exposure, or psychological harm.

Families should also distinguish entirely synthetic characters from manipulated images of real people. The Anderegg appeal involved images that prosecutors could not link to actual children.

That fact does not cover a classmate’s face placed onto a sexual image. It does not cover a family photograph altered into an explicit scene. Such content can identify and directly harm a real child.

A 2026 Pennsylvania case offers a concrete contrast. Two teenagers admitted involvement in creating about 350 fake nude images involving at least 59 girls under 18.

The source photographs reportedly came from yearbooks, social platforms, school pictures, and video chats. Victims described anxiety, lost trust, difficulty concentrating, and fear that the images would resurface.

The school deepfake case involved recognizable students. Its harms did not disappear because software helped create the images.

Another 2026 lawsuit accused xAI’s Grok image tools of transforming real photographs of teenagers into sexual images. The allegations include photographs taken from a homecoming event and a school yearbook.

That teen deepfake lawsuit remains an allegation, not a final finding against xAI. Still, it illustrates conduct outside the narrow facts considered in Anderegg.

These distinctions help families evaluate alarming claims without minimizing harm. A synthetic image can fall into several legal categories depending on its source, subject, location, and movement.

The relevant questions include whether a real child is identifiable, whether someone distributed the file, and whether it was sent to a minor. Investigators must also determine whether actual abuse material entered the creation process.

The Seventh Circuit AI ruling answers only part of that larger inquiry. It protects one form of private possession under the record before the court.

AI Has Undermined a Key Assumption From 2002

Modern generators have narrowed the factual gap between virtual images and records of real abuse, while constitutional doctrine still treats that gap as decisive.

When the Supreme Court decided Free Speech Coalition in 2002, realistic computer-generated people were much harder to create. The court still anticipated that visual technology would improve.

Justice Clarence Thomas warned that future tools might prevent prosecutors from proving whether images involved real children. Justice Sandra Day O’Connor raised similar concerns about computer-generated people resembling actual humans.

Those predictions now shape everyday investigations. Image models can produce photorealistic faces, bodies, lighting, and settings from short instructions. Local software can also operate without a platform reviewing each prompt.

Stable Diffusion is especially relevant because users can download model components and run modified versions on their own computers. A centralized service might never see those prompts or generated files.

This technical structure weakens platform-only safety strategies. A cloud provider can block certain requests, suspend accounts, or preserve evidence. Locally operated models give vendors less direct visibility and control.

The legal problem is even harder because “synthetic” does not always mean disconnected from real people. A generated image can be based on a real photograph, a fine-tuned model, or a collection containing unlawful material.

The Seventh Circuit’s concurrence noted research indicating that a major image-training dataset contained hundreds of known abuse images. The record in Anderegg was insufficient to resolve how training data affected the constitutional analysis.

That limitation matters. The court did not find that every synthetic image is created without exploitation. It found that the government conceded these charged images depicted no actual child and could not be linked to one.

The National Center for Missing and Exploited Children argues that even fully synthetic material burdens investigators. Each realistic report can consume resources needed to identify real victims.

NCMEC’s synthetic CSAM warning also says generated material can normalize abusive behavior and complicate victim identification. Those are policy and enforcement concerns, not automatic answers to a First Amendment case.

The government presented related arguments on appeal. It said synthetic material could stimulate demand, encourage offending, or make possession cases involving real victims harder to prove.

The panel concluded that Free Speech Coalition had already rejected substantially similar theories on the available record. It could not replace Supreme Court reasoning with its own assessment.

That result reveals the core tradeoff. A broad ban can help investigators respond to synthetic abuse, but it can also criminalize expression produced without an exploited subject.

Courts require the government to justify restrictions within established First Amendment categories. Technology can change the facts more quickly than appellate doctrine changes the test.

This gap pressures Congress, prosecutors, model developers, and the Supreme Court. Each institution controls only part of the response.

Congress can draft narrower statutes tied to identifiable harms or prohibited conduct. Prosecutors can build records showing how synthetic material affects investigations and real victims. Developers can restrict abusive outputs and preserve evidence where legally appropriate.

Only the Supreme Court can revise or clarify its own constitutional precedent for lower federal courts nationwide. That is why Judge Lee’s concurrence deserves more attention than a simplified Google News alert.

Families Face Harms the Possession Ruling Does Not Resolve

Families should treat the ruling as a narrow legal boundary, not as evidence that synthetic sexual abuse lacks victims or consequences.

An image can harm a child even when a court cannot prove the image records physical abuse. A recognizable face, bedroom, uniform, or school setting can make fabricated content personally targeted.

Distribution can deepen that harm. Copies can move between private chats, social networks, anonymous forums, and school communities. Removing one upload does not guarantee that every copy disappears.

Children may also face disbelief because the content is synthetic. A victim can still suffer humiliation, threats, harassment, or social isolation when classmates recognize the depicted person.

The Anderegg decision does not erase laws addressing stalking, harassment, extortion, unlawful disclosure, grooming, or nonconsensual intimate imagery. The exact options depend on the jurisdiction and the facts.

Federal enforcement has also continued in cases involving different conduct. In February 2026, the Justice Department announced a conviction involving receipt and possession of real abuse videos alongside obscene AI-generated material.

The federal conviction arose outside the Seventh Circuit and included facts unlike Anderegg. It does not settle the constitutional question, but it shows that prosecutions continue.

Parents should avoid treating every upsetting image as the same legal object. Preserving evidence can be more useful than debating labels before reporting it.

Useful records can include the original message, account name, platform, time, URL, and visible distribution history. Families should avoid forwarding the material except when law enforcement or an authorized reporting process instructs them.

A child depicted in fabricated content needs support even when no physical contact occurred. The emotional impact can resemble other forms of sexual exploitation and public humiliation.

Schools also need procedures that separate evidence preservation from unnecessary recirculation. Staff members should not create more copies than an investigation requires.

Platforms face a parallel responsibility. They need reporting paths that recognize synthetic or manipulated sexual content involving minors. A generic harassment category may not trigger appropriate review or preservation.

AI developers need safety tests for both text-to-image generation and image editing. Blocking a list of words is insufficient when users can employ indirect phrasing, reference images, or modified local models.

However, product controls cannot replace legal precision. Overbroad automated filters can block benign family photographs, health education, or abuse-prevention material.

That is the skeptical point missing from many reactions. A broader criminal rule might serve legitimate protective goals, but its wording still must distinguish unlawful conduct from protected expression.

The court did not say Congress lacks every option. It said this possession charge, under this record and existing Supreme Court cases, crossed the constitutional line.

Families reading the story through Google News should therefore resist two conclusions. The ruling neither makes all AI-generated abuse imagery lawful nor solves the enforcement problems surrounding it.

Three Signals Will Determine What Happens Next

The next phase depends on Supreme Court review, the remaining prosecution, and stronger evidence connecting synthetic material to identifiable harm.

The first signal is whether the Justice Department asks the Supreme Court to review United States v. Anderegg. A petition would give the justices an opportunity to reconsider how Stanley and Free Speech Coalition apply to modern generators.

Supreme Court review is not automatic. If the Court declines the case, the Seventh Circuit decision remains controlling within its region while national uncertainty continues.

If the justices accept the case, the key question will be narrow. They must decide whether private possession retains constitutional protection when synthetic imagery has become highly realistic.

A decision preserving the Seventh Circuit’s approach would strengthen the boundary between images involving real children and entirely virtual content. A reversal would expand government authority over in-home possession.

The second signal is what happens to the three remaining counts against Anderegg. Those proceedings will test the line between possession and outward conduct using the same underlying technology.

A conviction on production, distribution, or transfer charges would reinforce the ruling’s limited reach. A broader dismissal would increase uncertainty over how Section 1466A applies to locally generated material.

Those proceedings could also clarify relevant technical evidence. Courts may examine prompts, file histories, message records, software installation, and the alleged transfer to a minor.

The third signal is whether prosecutors develop a stronger factual record about harm from synthetic material. The Seventh Circuit repeatedly emphasized the constraints of the record before it.

Future cases might include evidence that a model trained on real abuse images. They might show that defendants used photographs of identifiable children or disguised real material as synthetic output.

Investigators could also document measurable burdens created by realistic false leads. That evidence would not guarantee a different constitutional result, but it would sharpen the government’s argument.

Congress may respond as well. A carefully written law could focus on production, solicitation, distribution, targeting, or transfer rather than private possession alone.

Any legislative proposal will face the same design challenge. It must protect children and support enforcement without treating every artificial depiction as proof that a real child was abused.

For families, the immediate action is simpler. Read beyond the headline, preserve evidence, use specialized reporting channels, and take a depicted child’s distress seriously.

Google News will keep compressing complex rulings into short alerts. The responsible question is not whether one alarming summary sounds permissive. It is which conduct the court addressed, which charges remain, and whether a real child was targeted.

Parents, schools, and platform teams should review their response plans before a case appears in their community. Who preserves evidence, who contacts authorities, and who supports the child?

Those decisions do not require waiting for another appellate ruling. They require treating synthetic sexual abuse as a real safety issue while keeping the law’s boundaries accurate.