AI & CSAM: It's not just AI, folks!

The Ruling Everyone's Furious About — But Aimed at the Wrong Target

If you've been on social media this week, you've probably seen the headlines: a federal court just said AI-generated child sexual abuse material is "legal." People are (rightly) horrified. But like most legal news that goes viral, the actual ruling is narrower, older, and more frustrating than the headlines suggest — and understanding why matters if you want to argue about it effectively instead of just yelling into the void.

The Outrage

"Using AI means you support CP"

If "AI enables this, therefore AI is complicit" were a real standard, it would apply to literally every image-making tool ever invented. It isn't a real standard. It's a red herring, deployed deliberately by people already at war with AI, who found in this ruling the perfect vehicle to make people angry at the technology itself. Child sexual abuse material is about as radioactive a topic as exists — attach it to "AI" in a headline and you don't need an argument anymore, you just need outrage. That's not concern for children. That's using children as a prop in a fight that was never really about them.

Consider this: if the above logic were true, then if you use cameras, you support CP, because cameras have been used to photograph it. If you use pen and paper, you support CP, because it's been used to draw it. If you take videos, you support CP, because it's been filmed. If you use the postal service, you support CP, because it's been mailed through it. If you use the internet, you support CP, because it's been uploaded, downloaded, and distributed across it. If you use digital art software, you support CP, because it's been used to illustrate it.

— which conveniently lets people skip the much harder, much older question of why the underlying legal doctrine has been sitting there, unfixed, for 25 years.

The facts, correctly stated

The ruling everyone's mad about — United States v. Anderegg¹ — didn't invent a new loophole. It applied a Supreme Court precedent from 2002, Ashcroft v. Free Speech Coalition², which struck down a federal law banning computer-generated images of minors in sexual conduct when no real child was used to create them., which struck down a federal law banning computer-generated images of minors in sexual conduct when no real child was used to create them. That's not a typo: 2002. Computer-generated imagery. Digital art, CGI, and rendering tools were well established by then — this wasn't some hypothetical future technology the Court couldn't have anticipated. The category already existed, and the Court ruled on it directly.

What changed between 2002 and now isn't the legal category. It's the tooling. The Seventh Circuit didn't create new legal ground — it applied the same doctrine to a defendant who happened to use Stable Diffusion instead of a 3D rendering program. The panel said so themselves, explicitly, while expressing real discomfort about it and inviting the Supreme Court to revisit the whole doctrine.

So here's my actual point: don't blame the tool

"AI" is not the legal mechanism that makes this material protected in someone's home. A 25-year-old First Amendment doctrine is. If you're aiming your outrage at generative AI as a category — as if this is some brand-new moral crisis that artificial intelligence uniquely created — you're pointing the gun at the wrong target, and worse, you're letting the actual target off the hook.

The Offenders

CSA vs. CSAM — they're not the same conversation

Before going further, it's worth untangling two things people keep collapsing into one outraged blob: child sexual abuse (CSA) and child sexual abuse material (CSAM). They're related, but they are not the same population, the same crime, or the same policy problem — and confusing them is part of why the discourse around this ruling has been so muddled.

CSA — the actual abuse of actual children — is overwhelmingly committed by someone the child already knows. This is not primarily a stranger-danger or dark-web phenomenon — it happens in homes, at family gatherings, in youth organizations, mostly at the hands of people already inside a child's circle of trust.

Who are the CSA perpetrators? (U.S.)

• 91% of child sexual abuse perpetrators are male; 9% are female.³

• 93% of child sex abuse perpetrators are someone known to the victim — 59% are friends/acquaintances, 34% are family members; only 7% are strangers.³

• By race: 57% of convicted perpetrators are White, 16% Black, 10% Hispanic.⁴

• Average age of convicted offenders: 38. Over 50% are over age 30.³

• Only about 3% of CSA perpetrators were unknown to the child, in substantiated cases.⁵

• An estimated 63,000 children a year were CPS-confirmed victims of sexual abuse (2009–2013).⁵

• A child is sexually assaulted, on average, every nine minutes in the U.S.⁶

• 1 in 9 girls and 1 in 20 boys under the age of 18 experience sexual abuse or assault.⁷

Who are the CSAM offenders?

CSAM offenders — people who access, possess, produce, or distribute the material — are a separately studied population, and the research paints a genuinely uncomfortable picture precisely because it doesn't match the "predator" stereotype at all.

• Mostly white men, ranging widely in age, with high levels of education and employment.⁸

• Tend to be single, and tend to have non-biological children rather than their own.⁸

• Show no distinguishing socioeconomic red flags — if anything, the opposite: higher-than-average education, occupational status, and employment rates.⁹

• Many did not deliberately seek out CSAM at first — most reported encountering it inadvertently or growing curious after viewing legal pornography.⁹

• Roughly 40–50% have a documented history of contact sexual offenses against children — meaning a substantial minority to near-majority do not.¹⁰ ¹¹

• Risk factors most associated with also being a contact offender include: a prior history of violent offenses, having groomed a child online, having had physical contact with a child, and searching specifically for material depicting infants or toddlers.¹⁰

• Most CSAM users are never caught — little is known about the much larger population that goes undetected, since almost all available data comes from those who were charged or convicted.¹⁰

Here's the part that should worry you more than any AI headline: research suggests meaningful overlap between the two groups, but not total overlap. Researchers have also noted that contact abuse can stem from multiple different causes — opportunity, power, antisociality, situational access — that don't necessarily involve a specific sexual attraction to children at all, whereas deliberately seeking out CSAM is harder to explain as anything other than that specific attraction. That distinction matters for treatment and risk-assessment purposes, but it should never be mistaken for minimizing contact abuse itself: a person who has sexually abused a child has caused real, irreversible harm to a real child, full stop, regardless of what mix of factors drove them to do it.

So: two overlapping but distinct populations, both disturbingly "unremarkable" in demographic terms — not defined by poverty, marginality, or obvious red flags, and not, in either case, defined by who's using what AI tool. Which is exactly why "spare me your moral outrage" isn't a dismissal of the danger — it's an insistence that we look at where the actual danger has always lived, instead of at whichever technology is trending in the headlines this week.

Now the harder question: where was everyone in 2002?

This is the part that should make people uncomfortable. The legal gap the Seventh Circuit was "forced" to apply isn't new. It's been sitting there, fully operational, since the Supreme Court decided Ashcroft. Computer-generated child sexual abuse material that didn't depict a real child has been constitutionally protected from certain prosecutions for over two decades. Not theoretically — actually, as settled law.

What is our Legal System doing to protect our children?

This is the big one, isn't it? Let's look at what happened.

1996 — Where this all actually started

Before getting mad at 2026, you have to look at 1996 — because that's when Congress first tried to deal with this exact problem, decades before "AI" was a word anyone used.

Back in 1996, Congress passed a law called the Child Pornography Prevention Act (CPPA).¹² They were worried about something specific: computer technology was getting good enough to create fake images of children in sexual situations — images that looked so real, people wouldn't be able to tell they weren't actual photographs. Congress wrote that concern right into the law, warning about images "virtually indistinguishable... from unretouched photographic images of actual children."¹²

So Congress tried to close the loophole early. Their law said: it doesn't matter if a real child was involved or not — if an image looks like a kid in a sexual situation, it counts as illegal child pornography, period.

That sounds like exactly what everyone wants right now. The problem is, six years later, the Supreme Court said that law went too far — and struck a big chunk of it down.

2002 — The Supreme Court says "too far"

In 2002, the Supreme Court took up a case called Ashcroft v. Free Speech Coalition, and it ruled that part of the 1996 law was unconstitutional.²

Here's the reasoning, in plain terms: the First Amendment protects freedom of speech, including a lot of speech people find disturbing or offensive. The Court said that the government can only ban material like this if it's either (a) made using a real child, which causes real, direct harm to that child, or (b) legally "obscene" under a separate, older test the courts already use for obscenity in general.

The 1996 law went further than that. It banned images just because they looked like a real child, even if the government couldn't prove any child was harmed and even if the image didn't meet the legal test for obscenity. The Court said that's too broad — you can't ban something just because it resembles an illegal thing, or because it might encourage bad behavior in someone who sees it. Under that logic, the Court said, you could end up banning things like mainstream movies or works of art that depict teenagers in sexual or romantic situations without literally using real minors.

So the Court struck down that part of the law. Not all of it — laws against material that actually depicts real children remained fully in place, no change there at all. But the part that banned purely fictional or computer-generated depictions, where no real child was used, got thrown out.

That's the precedent — a 2002 ruling — that later courts, including the one everyone's angry at in 2026, say they're legally required to follow.

Who comprised SCOTUS in 2002?

Majority (voted to strike down the CPPA provisions) — 6 justices

• Anthony Kennedy — wrote the majority opinion. Reasoned that a law banning material based only on the fear that it might encourage illegal acts goes too far; the connection between virtual images and actual child abuse is "contingent and indirect," and "the mere tendency of speech to encourage unlawful acts is not a sufficient reason for banning it."²

• John Paul Stevens — joined the majority opinion in full.²

• David Souter — joined the majority opinion in full.²

• Ruth Bader Ginsburg — joined the majority opinion in full.²

• Stephen Breyer — joined the majority opinion in full.²

• Clarence Thomas — concurred in the judgment (agreed with the outcome, joining the "majority" numerically) but wrote his own separate opinion rather than fully joining Kennedy's reasoning.²

Concurrence

• Clarence Thomas — agreed the law as written was unconstitutional, but wrote separately to flag a specific worry: that advancing technology could soon make virtual images "indistinguishable" from real ones, which might justify a different, more carefully drafted law in the future. This is the passage now being cited by courts and commentators in 2026 as remarkably prescient.²

Dissent (voted to uphold the CPPA provisions, at least in part) — 3 justices

• William Rehnquist (Chief Justice) — dissented, arguing the Court should have narrowed the law's interpretation to save it rather than striking it down entirely.²

• Antonin Scalia — joined Rehnquist's dissent (except for one paragraph discussing legislative history).²

• Sandra Day O'Connor — dissented in part. She agreed that banning "youthful-adult" pornography (real adult actors made to look underage) went too far, since it's evident to a viewer no real minor was used. But she argued Congress could still constitutionally ban purely computer-generated/virtual child pornography specifically, treating it as its own, more dangerous category — a distinction the majority rejected.²

  
  

The interesting wrinkle here: this wasn't a partisan split at all. Of the 6 justices in the majority, 4 were appointed by Republican presidents (Kennedy, Stevens, and Thomas via Reagan/Ford/Bush) and only 2 by a Democratic president (Ginsburg, Breyer via Clinton). All 3 dissenters were also Republican appointees. So the doctrine that's driving the 2026 outrage wasn't handed down by some ideological faction — it was written and controlled by justices appointed across both parties, with Republican appointees making up the bulk of both sides of the vote.

2026 — Same rule, new technology

Fast forward to this year. A man named Steven Anderegg used an AI image generator called Stable Diffusion to create hundreds of sexually explicit images of children.¹ None of the images were of a real child — they were entirely computer-generated, built from a text description he typed in.

He was charged with several crimes: producing these images, distributing them, sending them to a minor, and simply possessing them at home. A lower court looked at just one of those charges — the possession charge — and threw it out, using the same reasoning from 2002: if no real child was used to make the image, and it doesn't meet the legal test for obscenity, then banning someone from privately having it in their own home violates the First Amendment.

The government appealed, asking a higher court — the Seventh Circuit Court of Appeals — to reverse that decision. In August 2026, that court said no.¹ Three judges looked at the case and said, in effect: "the 2002 ruling still applies here. An AI-generated image with no real child in it is legally the same category the Supreme Court already ruled on. We don't like it, but we don't get to overrule the Supreme Court — only the Supreme Court can do that."

Importantly, the judges didn't say they were happy about it. They pointed out that the technology available today is far more realistic and far easier to use than anything that existed when the Supreme Court decided this in 2002 — and they specifically said they think the Supreme Court should take a fresh look at the issue, given how much has changed.

And to be clear about what didn't happen: Anderegg is still being prosecuted for producing these images, distributing them, and sending them to a minor. Only the "having it privately at home" charge was affected by this ruling.

The real shame

Here's what should actually outrage you: this gap in the law isn't new. It's been sitting there since 2002. For over two decades, purely computer-generated images of children in sexual situations — made with regular digital art tools, long before AI — have been protected from this specific kind of prosecution in the home. That should have been a scandal in 2002. It should have stayed a live issue in the years since. Where was the outrage then? Where has it been for twenty-three years while this doctrine just sat there, unfixed, unnoticed, ignored?

And here's the uncomfortable irony: the only reason any of us are talking about this now is because a small, vocal minority of people who despise AI — for reasons that have nothing to do with child safety — decided to seize on this ruling as their weapon. They didn't dig up a 25-year-old legal doctrine because they suddenly cared about protecting children. They found a ruling with "AI" in the headline and the word "child" nearby, and they ran with it, because outrage bait doesn't get better than that.

So, strange as it is to say: we owe them a small, bitter thank-you. Not because their motives were honest, but because their dishonesty accidentally did something worth doing — it dragged genuinely broken laws out into the daylight, twenty-three years late, for entirely the wrong reasons.

Where the anger should actually go

If you want to be angry — and this is a genuinely maddening situation — aim it somewhere that can fix it:

• Congress, which could write a narrower, more precise statute addressing this material in a way built to survive the First Amendment scrutiny that sank the 1996 law.

• The Supreme Court, which the Seventh Circuit itself all but begged to revisit Ashcroft given how far the technology has come.

• The AI companies and dataset curators whose training data has reportedly contained actual CSAM — a real, current, fixable problem that two of the three judges flagged directly in their concurrence, and that has nothing to do with the constitutional question at all.

None of those targets are as satisfying to yell at as "AI is evil." But this is where things need to be changed. Not just for AI-generated CSAM but for all CSAM. All of it. "Real child" or not "real child," but any depiction. The doctrine has been broken for 25 years. AI didn't break it. AI just showed us that we, as a society, have a huge problem that needs to be addressed.

Sources

1. United States v. Anderegg, No. 25-1354 (7th Cir. Aug. 25, 2026).

2. Ashcroft v. Free Speech Coalition, 535 U.S. 234 (2002).

3. Lawsuit Information Center, "Child Sex Abuse Statistics," compiling government/nonprofit data, updated May 18, 2026. lawsuit-information-center.com/child-sex-abuse-statistics.html

4. U.S. Sentencing Commission data, as cited in source 3.

5. RAINN, "Statistics: Children & Teens," citing U.S. Department of Health and Human Services, Administration for Children and Families, Administration on Children, Youth and Families, Children's Bureau, Child Maltreatment report, Exhibit 5-2, data years 2009–2013. rainn.org/statistics/children-and-teens

6. RAINN, "Statistics: Children & Teens," citing U.S. Department of Justice, Office of Justice Programs, Bureau of Justice Statistics, National Crime Victimization Survey, 2020–2024 average.

7. RAINN, citing David Finkelhor, Anne Shattuck, Heather A. Turner, & Sherry L. Hamby, "The Lifetime Prevalence of Child Sexual Abuse and Sexual Assault Assessed in Late Adolescence," Journal of Adolescent Health 55 (2014): 329–333. rainn.org/node/2958

8. Systematic review of 35 studies on CSAM offender risk profiles and typologies, PubMed, published January 2026.

9. "Accessing child sexual abuse material: Pathways to offending and online behaviour," ScienceDirect, 2024.

10. "Investigating the disparities among child sexual abuse material users: Anonymous self-reports from both charged and uncharged individuals," ScienceDirect, 2025.

11. "Unmasking the Men Who Produce Child Sexual Abuse Material (CSAM): Criminal Diversity and Sociodemographic Characteristics," PMC, 2025.

12. Child Pornography Prevention Act of 1996, Pub. L. No. 104-208, Div. A, Tit. I, § 121, 110 Stat. 3009 (1996).