Forum: Author Hangout

Revenge porn is a crime in 48 states.

The definition of 'revenge porn' in many of those states does not equate to these cases, though. Yes, I'm aware you didn't say it does :)

For instance, under Texas law:

(2) the visual material was obtained by the person or created under circumstances in which the depicted person had a reasonable expectation that the visual material would remain private;

is an element of the crime. In the case of 'hybrid' works, the depicted person cannot have any reasonable expectations since they weren't even there.

We're getting there, but we're not there yet.

Probably a series of pictures of a naked 13-year-old girl, of girls.

If a given AI has been trained using illegal content, then the person who fed that content into the AI, and possibly their whole company, might be charged under distribution laws.

That depends. Existing distribution laws require actual knowledge of the content, not merely suspicion (and there is no 'they should have known' exception, either).

It's pretty certain that LAION-5B contained some amount of alleged CSAM, but no one is certain what that CSAM might be, and it might not be illegal. All that is known is that the checksum hash values of some of LAION-5B's images match the hash values of images in a CSAM database. No one has the images handy (for fairly obvious reasons, in the case of the database).

Mind you, LAION-5B is a set of links and text descriptions. It 'contains' no images. Those who made it available never viewed the images - descriptions were created by AI. Those who used it never viewed the images. There were almost six billion images in LAION-5B. How could someone(s) possibly have audited them, especially a small nonprofit?

I can see this logic applying even if the AI was simply hooked up to a spiderbot and instructed to find images. Given that the AIs are not considered people under the law, a person is in charge of them legally. That person is responsible for what the AI consumes.

Again, the law requires 'knowingly'. No one 'knowingly' did anything.

Similarly, I suspect there are already major cases working their way through the system that are going regarding intentional copyright infringement due to AIs being trained using material that's under copyright. Again, doesn't matter how the AI learned, someone is ultimately responsible for the AI.

There are. The odds are high that it'll be found to be fair use, but if not, no one is discussion criminal liability, only licensing agreements.

In Japan, there's a whole genre where the core advertising number of a pornstar's bio is her weight (the less the better) next to height (same).

Isn't the age of 'consent' also shockingly low in Japan...

AI tools are pretty good at 'realism' filters also - taking cartoon to real and back. So if you get something that was a cartoon in Japan, it can figure out what the realism of that would be, and other filters can alter the ethnicity.

That noted. Most of the stuff I've seen on AI training galleries is either American or Chinese. The "Japanese stuff" is usually actually Chinese. Even the Hentai / XXX. And Chinese stuff might sexualize women a LOT, but not children.

Apparently they raised it to 16 last year (it was 13) but even that wasn't the lowest in the world.

It was a bit more complicated, as intercourse was technically illegal for anyone under eighteen except in 'legitimate' romantic relationships (not clearly defined, but effectively if neither set of parents objected).

The AI knows what 13yo girls look like and it knows what a naked adult female look like. From there it's just a matter of interpretation.

This seems very similar to the case that Switch Blayde quoted:

They created and shared images that superimposed pictures of real students' faces onto simulated nude bodies generated by artificial intelligence.

In both cases, it's an AI-abetted deepfake. The original article mentions 'child pornography,' but only in a 'the law doesn't apply' sense.

One is an expulsion, the other is criminal charges, but both are about AI deepfakes of real people, not 'virtual child pornography' in particular.

The California law might actually apply - it's arguable that it would cover an AI-generated image that used an actual person's image as an input. If they just generated a million images at random and stopped when one looked like a classmate, that would be different. In either case, as the original article states, it would be a difficulty for prosecutors.

Oh, I agree with you there, except that it is considerably easier for the average person to make a photorealistic output using AI than with PhotoShop. Despite Adobe's claims, 'old school' Photoshop isn't that easy, and 'newfangled' Photoshop is AI.

Also, a 'photo chop' with 12-year-old heads and 18-year-old bodies is probably a Frankenstein mess. Getting 12-year-old bodies makes it very questionable. No necessarily 'child pornography', because mere nudity is not always consider CSAM, but it's an entirely different thing than using AI to generate a fake 12-year-old body.

My suspicion was that DBActive was referring to the PROTECT Act of 2003, not Mike Lee's PROTECT Act of 2024.

https://en.wikipedia.org/wiki/PROTECT_Act_of_2003

DBActive may be referring to:

Prohibits computer-generated child pornography when "(B) such visual depiction is a computer image or computer-generated image that is, or appears virtually indistinguishable from that of a minor engaging in sexually explicit conduct"; (as amended by 1466A for Section 2256(8)(B) of title 18, United States Code).

That applies to obscenity (which Miller does not cover). It would not apply to a 'real child, face / fake body' case if there was no sexual activity, 'lascivious exhibition of the genitals or pubic area', etc. It wouldn't matter if the child themselves weren't doing anything sexual - the image as a whole represents a child engaged in sexual activity, which is illegal.

Note that the PROTECT Act of 2003 has been tested in court with respect to traditional animation (so, not photorealistic) of children engaged in sexual activity. Since that made it obscene under Miller (and thus not covered by Ashcroft), the person was convicted. As the Wikipedia entry notes:

Obscenity, including obscene depictions of children, either virtual or real, is unprotected speech

There is some belief that parts of that act remain unconstitutional, and prosecutors have made plea deals to avoid raising those constitutional challenges, but that's hypothetical.

I'd want to see the actual cases. I would be willing to bet that there was more going on than just children engaged in sexual activities.

From Wikipedia (yes, not the best source, but it covers a fair bit of what you're asking):

The first conviction of a person found to have violated the sections of the act relating to virtual child pornography was Dwight Whorley of Virginia, who used computers at the Virginia Employment Commission to download "Japanese anime style cartoons of children engaged in explicit sexual conduct with adults" alleged to depict "children engaged in explicit sexual conduct with adults". He was charged with 19 counts of "knowingly receiving" child pornography for printing out two cartoons and viewing others. His conviction was upheld in a 2–1 panel decision of the Fourth Circuit Court of Appeals in December 2008. This decision was consistent with the U.S. Supreme Court ruling in Ashcroft v. Free Speech Coalition in which the Supreme Court held that virtual child pornography was protected free speech, provided that the virtual depictions are not obscene. Obscenity, including obscene depictions of children, either virtual or real, is unprotected speech. (Whorley was also previously convicted of offenses in connection with pornographic depictions of real children.)

The case is: https://caselaw.findlaw.com/court/us-4th-circuit/1431669.html

There are some potentially disturbing things in the Fourth Circuit opinion in terms of e.g. SoL:

We also reject his arguments that textual matter cannot be obscene under § 1462 and that cartoons depicting minors in sexually explicit conduct must depict real-life minors to violate § 1466A(a)(1).   Finally, we reject his challenges to the district court's procedural rulings and his sentence.   Accordingly, we affirm.

Note the part about textual material. Most SoL stories would pass Miller with no problem, but an 'All Sex' might not.

The case clearly involves only textual and cartoon material, but yes, you are correct that there is violence, rape, etc. The point is more that it's clear that visual and textual material which do not involve actual children can be held obscene and can be punished under existing law, at some level of obscenity (yet to be determined).

I agree that not everything depicting sex involving minors is obscene, but the 2003 PROTECT act effectively defines it that way, and (to my knowledge - could be wrong) that has not been tested in court.

There's an important caveat in US vs Williams:

The Court further stated that 18 U.S.C. § 2252A(a)(3)(B) would not be construed to punish the solicitation or offering of "virtual" (computer generated/animated) child pornography, thus comporting with the holding of Ashcroft v. Free Speech Coalition, 535 U.S. 234 (2002).

The Court stated that "an offer to provide or request to receive virtual child pornography is not prohibited by the statute. A crime is committed only when the speaker believes or intends the listener to believe that the subject of the proposed transaction depicts real children. It is simply not true that this means 'a protected category of expression [will] inevitably be suppressed,' post, at 13. Simulated child pornography will be as available as ever."

Thus, computer generated/animated CSAM is likely still covered by Ashcroft unless a court finds it to be obscene (in which case Ashcroft would not necessarily apply). Williams explicitly does not change that, and the PROTECT act of 2003 thus does not bar virtual CSAM under the pandering provision, per current precedent.

Metadata is the answer to your first point. Tag the image with the metadata used to recreate it algorithmically. Done.

For the second point, if one merely need send the prompt and setup, no images need be traded. And if one cannot be prosecuted for virtual images, who would trade for real images? Yes, that's a somewhat hyperbolic statement. There are people who would only be satisfied by real images. But it quite likely creates a major divide in which a large number of people who don't children to be harmed can opt out of it, while those who do wish children to be harmed essentially wave their hands in the air and acknowledge that. That, in turn, tells people whose concern is the protection of children which body of people to focus on and removes a large source of distraction: people who aren't a threat to children and whose prosecution doesn't make children safer.

On your third point, yes, if it's primarily an image of a real child. 'Child head on adult body' where the body 'looks young' but is provably adult, especially where there's no obscenity, seems very likely to be covered under Ashcroft. It may run afoul of 'deepfake' laws, though.

Part of the text of the PROTECT act of 2003 fits exactly your first point (that everyone will claim it's virtual, and it'll be a bottleneck to prosecutions to force the government to prove that it's not). Nevertheless, the Williams court explicitly, in so many words, stated that the 'pandering' provisions do not apply to virtual child pornography. If they're not going to apply the pandering provisions, why would they apply possession or distribution?

Repeating for emphasis:

A crime is committed only when the speaker believes or intends the listener to believe that the subject of the proposed transaction depicts real children.

That's a pretty strong statement.

But, the images (child's face on a virtual body) does depict a real child.

I disagree. The images depict something which is definitely not a real child. It's part of a real child and part not, which makes it not a real child.

But, even if I grant that point for argument, that's still not responsive to the quoted text. The quoted text says that either 'the speaker believes' or 'intends the listener to believe'

The speaker does not believe the subject depicts real children.
The speaker does not intend the listener to believe the subject depicts real children.

Thus, there is, per the Williams court, no crime.

An actual child is depicted in a sexually explicit manner.

There is no 'actual child' here. That was my entire point. By definition, this is not a depiction of a real child engaged in the behavior shown. Everyone knows that going in. That's a fact.

'The point of the law is to protect children. Creation and distribution of realistic images of a child in a sexually explicit situation harms the child depicted.'

Only if the image is of a specific child and is believed to be that child. Otherwise, no. What you're arguing for here is a 'deepfake' law that doesn't exist and, even if it did, functions differently than nearly all 'deepfake' laws.

I hate to keep banging the same drum, but you're not hearing it. The Williams court was extremely explicit: if the speaker does not believe a real child is depicted (and they clearly do not, since the image is factually not the image of a real child) and the recipient is notified that this is not the image of an real child, there is no crime. Period.

Based on all the cases I cannot believe the courts would give those images a pass.

You are welcome to believe that, just as people for years believed some court, somewhere, would overturn Roe, and eventually one did. But, prior to that, the state of the law was what it was. As of right now, we have a clear statement of the law, and it is that the 'pandering' rule does not apply when neither the distributor or the recipient believe the image is that of an real child.

Dropping children from this, because they honestly are irrelevant to the analogy: consider the recent Taylor Swift (TS) deepfake mess.

If:
1) The poster says 'These are actual images of TS! Look what she's doing!' that may be actionable under some 'deepfake' laws, and there may (or may not) be grounds to sue under other laws.

2) The poster says nothing, just puts out the images, that is less likely to be actionable under either (but non-zero).

3) The poster says 'These are not images of TS, TS did not do these things, and this is just a parody', very few 'deepfake' laws apply and the other laws also very likely don't apply.

What you seem to be arguing is that, even though the poster says 'this is not X', and the recipient says 'this is not X', and both believe that, and also the image factually is not X, still somehow the image is X. That makes no sense, and I would be very worried if any court were to make such a ruling, because once a court rules that an image no one believes to be real and that is also factually not real is yet to be treated legally as if it were real, all bets are off.

Creation and distribution of realistic images of a child in a sexually explicit situation harms the child depicted.

The man on the Clapham Omnibus certainly wouldn't want pictures of his kids used in that way. I believe, in the UK, privacy laws means the child's face cannot be used unpixelated without permission.

AJ

Very different SCOTUS, though. I have no confidence that the current court will respect precedent in general But, then, nor do I think they would necessarily overturn, just that I don't see a good way to predict this one this far out.

You're wrong. The statute was amended in 2022

Not me. The L.A. Times. I copied from their article dated March 7, 2024.

Can simple nudity be classified as porn? I would think it needs to portray sex.

I remember years ago a guy was charged with child porn because he had a nude photo of his kid in the bathtub or something like that. I don't know if it went to trial or, if it did, if he was convicted. I doubt it.

But the reason for my post wasn't about the law. It was about using AI.

Can simple nudity be classified as porn?

In the US, 'simple nudity' is generally not 'porn' unless the pose is considered so sexualized as to make the photo one taken for erotic purposes. Photos of people (adults, kids, both) just standing around, playing, eating, whatever: pretty much legal. Someone (regardless of age) lying back on a bed, legs spread suggestively: probably porn.

For the quintessential example involving children, check the work of David Hamilton. His photography of naked children is legal in the US.

It's likely illegal in Canada, on the other hand.

I know of Brits who purchased Hamilton books from Amazon.co.uk. It would be weird if that were illegal.

AJ

There are also libraries which have every issue; no one has gone after them.

The flip side of this is that some attempts have been made to go after people both for sending and for receiving naked selfies of teenagers. By sending, I don't mean resending - there are at least been allegations made that a person sending a naked selfie of themself commits a crime in so doing and that having that picture on their phone is itself a crime even if it was never sent.

I suspect they won't hold up in court, but it's a place where the law is running headlong into 'unexpected consequences'. Redistributing a selfie without consent may be both wrong and possibly criminal, but it's not what CSAM laws were designed to deal with, and having a selfie of oneself is very much something those laws did not foresee.

There have been cases of minors being placed on the sex offender registry for sharing nude selfies of themselves or others, though so far they've all been juvenile offenses. However, in at least one of the cases, the offender will be listed as a sex offender into his 40s.

There's a case where two girls wearing bras swapped selfies and were threatened with prosecution (but not prosecuted).

No dispute that there are legitimate and reasonable consequences. Expulsion seems reasonable in limited circumstances (non-consensual sharing, for instance, which might also warrant criminal and civil penalties).

Trying them as 'child porn', though (even when prosecuted as a juvenile) or putting someone on a decades-long sex offender registry seems excessive, though.

Most 'deepfake' legislation requires the image to be distributed, and quite a bit of it requires some level of misrepresentation of the image as being that of the target.

That's not true of all of the laws, but for most of them, keeping the images to yourself is legally fine. Similarly, distributing the images by saying 'These are definitely not X, Y, or Z, nope, not way, not them at all' would also steer clear of the law.