“It’s time to give victims their day in court and the tools they need to fight back,” says Sen. Dick Durbin as the DEFIANCE Act heads to the House.
THE SENATE UNANIMOUSLY passed a bipartisan bill to provide recourse to victims of porn deepfakes — or sexually-explicit, non-consensual images created with artificial intelligence.
The legislation, called the Disrupt Explicit Forged Images and Non-Consensual Edits (DEFIANCE) Act — passed in Congress’ upper chamber on Tuesday. The legislation has been led by Sens. Dick Durbin (D-Ill.) and Lindsey Graham (R-S.C.), as well as Rep. Alexandria Ocasio-Cortez (D-N.Y.) in the House.
The legislation would amend the Violence Against Women Act (VAWA) to allow people to sue those who produce, distribute, or receive the deepfake pornography, if they “knew or recklessly disregarded” the fact that the victim did not consent to those images.
Once it gets to a vote, yes. The way to kill these things is to bury it in procedure and pretend you're just doing due diligence. Still pretty amazing that it got this far.
Things have already been moving towards nobody models. I think this will eventually have the consequence of nobodies becoming the new somebodies as this will result in a lot of very well developed nobodies and move the community into furthering their development instead of the deepfake stuff. You'll eventually be watching Hollywood quality feature films full of nobodies. There is some malicious potential with deep fakes, but the vast majority are simply people learning the tools. This will alter that learning target and the memes.
That's where the money is, so yes that's where the majority of work is. But I do think one of the drivers of this is to help protect more local instances; to create consequences for things like fake revenge porn or distributing deepfakes of classmates/teachers in your school, etc.
This might make the path to generating slightly harder, but it won't do anything to stop an intelligent person. I haven't seen a ton of info from people talking about this stuff, but exploring on my own, especially with Stable Diffusion 3, diffusion models are very different than LLM's. The filtering for safety type alignment is happening external to the model using CLIP, and in the case of SD3, 2× CLIP models and a T5xxl LLM model. The alignment filters are done with these and some trickery. Screwing with these can enable all kinds of capabilities. It is just hard to understand the effect of some tricks, like SD3 swaps an entire layer in the T5 manually. When these mechanisms are defeated, models can generate freely, which essentially means everything is a deepfake. This is open source. So it can never be extinguished. There was a concerted effort to remove the rogue 4chanGPT. It does not have the ChatGPT derived alignment like all other models. The 4chanGPT is still readily available if you know where to look.
This bill just raises the barrier of entry and makes such content less familiar and more powerful in the end. In reality, we would be socially stigmatizing while accepting the new reality IMO. This is like an weapons arms race. You may not like that the enemy created cannons, but banning the casting of cannons within your realm will do nothing to help you in the end. Everyone in the realm may understandably hate cannons, but you really need everyone familiar with casting, making the and everyone in your realm to learn how to deal with them and what to expect. The last thing you need is a lot of ignorant people on a battlefield bunching up together because they do not understand their opponents.
These tools are also weapons. Everyone needs to understand what is truly possible, regardless of how unpleasant that may seem. They can not have a healthy skepticism without familiarity. If they do not have familiarity, they will bunch up on a battlefield facing cannons loaded with grapeshot.
Of course, eventually somebody will look exactly like the nobody, so the owners of the nobody will sue that somebody to block them from pretending to be that nobody in videos.
I see it eventually happening in porn, but it isn't happening in major motion pictures any time too soon. People won't follow and actively go see a cgi actor in a movie like they would a real one. Hollywood pays big money for A list stars because those people get asses in seats. That, along with tech not being there quite yet, and Hollywood locked under sag contract to only do certain things with AI all adds up to nobody's being a ways off.
I also remember years ago reading scare headlines about how the CG in Final Fantasy: The Spirits Within was so realistic that it would be the end of Hollywood actors, they would all be CG.
So I'll take that claim with a huge grain of salt.
As always, I read the bill expecting to be deeply disappointed; but was pleasantly surprised with this one. It's not going to solve the issue, but I don't really know of anything they can do to solve it. My guess is this will mostly be effective at going after large scale abuses (such as websites dedicated to deepfake porn, or general purpose deepfake sites with no safeguards in place).
My first impressions on specific parts of the bill:
The bill is written as an amendment to the 2022 appropriations act. This isn't that strange, but I haven't actually cross-references that, so might be misunderstanding some subtlety.
The definition of digital forgery is broad in terms of the means. Basically anything done on a computer counts, not just AI. In contrast, it is narrow in the result, requiring that:
when viewed as a whole by a reasonable person, is indistinguishable from an authentic visual depiction of the individual.
There is a lot of objectionable material that is not covered by this. Personally, I would like to see a broader test, but can't think of any that I would be comfortable with
The depiction also needs to be relevant to interstate or foreign commerce. There hands are tied by the constitution on this one. Unless Wickard v Fillburn us overturned though, me producing a deepfake for personal use reduces my interstate porn consumption, so it qualifies. By explicitly incorporating the constitutional test, the law will survive any change made to what qualifies as interstate commerce.
The mens rea required is "person who knows or recklessly disregards that the identifiable individual has not consented to such disclosure" No complaints on this standard.
This is grounds for civil suits only; nothing criminal. Makes sense, as criminal would normally be a state issue and, as mentioned earlier, this seems mostly targeted at large scale operations, which can be prevented with enough civil litigation.
Max damage is:
$150k
Unless it can be linked to an actual or attempted sexual assult, stalking or harassment, in which case it increases to $250k
Or you can sue for actual damages (including any profits made as a result of the deepfake)
Plaintifs can use a pseudonym, and all personally identifiable information is to be redacted or filed under seal. Intimate images turned over in discovery remains in the custody of the court
10 year statute of limitations. Starting at when the plaintif could reasonably have learned about the images, or turns 18.
States remain free to create their own laws that are "at least as protective of the rights of a victim".
My guess is the "at least as protective" portion is there because a state suite would prevent a federal suit under this law, as there is an explicit bar on duplicative recovery, but I have not dug into the referenced law to see what that covers.
How close does the deep fake need to be to the original? What we saw with DALLE2 was that each person whose face was restricted made a hole in the latent space of all faces. After enough celebrities faces were restricted, there were so many latent space holes that the algorithm couldn't make faces at all since every producable face was a certain "distance" away from a restriction.
Sure, you can make lora training on unconsenting people illegal and also make particular prompts illegal, but there is enough raw data and vague prompts to mold a generation into something that looks like it was done the illegal way without breaking either of those two restrictions.
There are billions of people. Find the right one, and a "reasonable person" could not tell the difference.
Image a law that said you cannot name your baby a name if someone else's name starts with the same letter. After 26 names, all future names would be illegal. The law essentially would make naming babies illegal.
The "alphabet" in this case is the distict visual depiction of all people. As long as the visual innumeration of "reasonable people" is small enough, it essentially makes any generation illegal. Conversely, if "reasonable people" granulated fine enough, it makes avoiding prosecution trivial by adding minor conflicting details.
And what would be lost? I might be missing something, but what is the benefit of being able to make fake people faces that outweighs the damage it can do to people's lives and the chaos wrecked on society from deepfakes etc?
Maybe if society was more reasonable and responsible with attitudes towards sexuality then deepfakes and sexual crimes would naturaly not be significant issues.
if everyone had a box that gave them the sexual gratification they needed, then they could go about the rest of their day without injecting sexual wants into normal activities and relationships.
Adverts and marketing might have to use facts instead of sex to sell products.
I would be careful about assuming knowledge based on age. Young people might use technology without understanding it, and old people might understand it and don't want to use it.
Technology needs to be regulated, and I would not trust people with profit incentives to do so.
IMO, it is always important to investigate if a regulation wants to prevent a real issue or if they just mention some populist reasons for doing whatever they want.
Why even specifically mention AI? Are there already laws that cover creating a sexually explicit likeness without consent e.g. with photoshop or just painting one? If so why wouldn't those laws also cover AI and if not why wouldn't the law also wish to cover these cases?
I didn't post with an alternative solution in mind as much as I was looking to elicit conversation that provided more perspective and context that would assuage my concerns. -I've since got that from some of the others responses.
You need to start somewhere. Regulation will always lag behind technology. But sooner or later things will get regulated. Once a good number of people are affected by something, rules will be brought in by the people. That's how democracy works.
These rules are never perfect. Sometimes, to make rules effective, they have to be multilayered (swiss cheese model). But that makes them too expensive to implement. So eventually things end in a compromise where cost and effectiveness balance.
At least they have now started to try to learn about the tech they're trying to regulate, as opposed to Ted Stevens who obviously just read a prepared speech someone else had made in his famous "Series of Tubes" speech.
Ok, fist of all, politicians need to stop it with these acronyms for every law they want to pass. It’s getting ridiculous. Just give the damned law a regular-ass name. It doesn’t have to be all special and catchy-sounding damn.
Second, I’m really surprised to hear of anything passing the senate unanimously, other than a bill expressing the love of silly acronyms. And weak campaign finance laws.
Anyway, I’m glad that at least something is being done to address this, but I just know someone in the House is gonna fuck this up.
It passed by unanimous voice vote. That is, the president of the Senate (pro tempore, usually Patty Murray) asked for all in favor, and at least one person said "yea" and when asked for all opposed, nobody said "nay". There wasn't a roll call vote, so we don't know how many people (or who) actually voted for it.
Edit: it was probably on C-Span, so you can probably find a recording and get an idea of how many people were there, and how many yeas you hear, if you're so inclined.
I think so. In a way, it makes sense -- a lot of people are of the (shitty) opinion that if you take lewd pictures of yourself, it's your fault if they're disseminated. With lewd deepfakes, there's less opportunity for victim blaming, and therefore wider support.
To improve rights to relief for individuals affected by non-consensual activities involving intimate digital forgeries, and for other purposes.
Congress finds that:
(1) Digital forgeries, often called deepfakes, are synthetic images and videos that look realistic. The technology to create digital forgeries is now ubiquitous and easy to use. Hundreds of apps are available that can quickly generate digital forgeries without the need for any technical expertise.
(2) Digital forgeries can be wholly fictitious but can also manipulate images of real people to depict sexually intimate conduct that did not occur. For example, some digital forgeries will paste the face of an individual onto the body of a real or fictitious individual who is nude or who is engaging in sexual activity. Another example is a photograph of an individual that is manipulated to digitally remove the clothing of the individual so that the person appears to be nude.
“(3) DIGITAL FORGERY.—
“(A) IN GENERAL.—The term ‘digital forgery’ means any intimate visual depiction of an identifiable individual created through the use of software, machine learning, artificial intelligence, or any other computer-generated or technological means, including by adapting, modifying, manipulating, or altering an authentic visual depiction, that, when viewed as a whole by a reasonable person, is indistinguishable from an authentic visual depiction of the individual.
It seems like the bill is being pitched as protecting women who have fake nudes passed around their school but the text of the bill seems more aimed at the Taylor swift case.
1 The bill only applies where there is an “intent to distribute”
2 The bill talks about damages being calculated based on the profit of the defendant
The bill also states that you can’t label the image as AI generated or rely on the context of publication to avoid running afoul of this law. That seems at odds with the 1st amendment.
The bill only applies where there is an “intent to distribute”
That's a predicate for any law bound to the Commerce Clause. You need to demonstrate the regulation is being applied to interstate traffic. Anything else would be limited to state/municipal regulations.
The bill talks about damages being calculated based on the profit of the defendant
That's arguably a better rule than the more traditional flat-fee penalties, as it curbs the impulse to treat violations as cost-of-business. A firm that makes $1B/year isn't going to blink at a handful of $1000 judgements.
The bill also states that you can’t label the image as AI generated or rely on the context of publication to avoid running afoul of this law.
A revenge-porn law that can be evaded by asserting "This isn't Taylor Swift, its Tay Swiff and any resemblance of an existing celebrity is purely coincidental" would be toothless. We already apply these rules for traditional animated assets. You'd be liable for producing an animated short staring "Definitely Not Mickey Mouse" under the same reasoning.
This doesn't prevent you from creating a unique art asset. And certainly there's a superabundance of original pornographic art models and porn models generated with the consent of the living model. The hitch here is obvious, though. You're presumed to own your own likeness.
My biggest complaint is that it only seems to apply to pornography. And I suspect we'll see people challenge the application of the law by producing "parody" porn or "news commentary" porn. What the SCOTUS does with that remains to be seen.
That’s arguably a better rule than the more traditional flat-fee penalties, as it curbs the impulse to treat violations as cost-of-business. A firm that makes $1B/year isn’t going to blink at a handful of $1000 judgements.
No argument there but it reinforces my point that this law is written for Taylor swift and not a random high schooler.
You’d be liable for producing an animated short staring “Definitely Not Mickey Mouse” under the same reasoning.
Except that there are fair use exceptions specifically to prevent copyright law from running afoul of the first amendment. You can see the parody exception used in many episodes of south park for example and even specifically used to depict Mickey Mouse. Either this bill allows for those types of uses in which case it’s toothless anyway or it’s much more restrictive to speech than existing copyright law.
Question from an outsider:
Do all bills in the states have to have a fancy acronym?
It looks like the senate is the first step, is that right? Next is the house? It's the opposite where I am.
Not all bills do, but the majority of big ones you hear about do. It's simply a marketing thing
And correct, it'll move to the House, where if it passes it will move to the president's desk. Considering it was unanimous in the Senate, I can't see it having any issues in the House.
Bills in the US can originate from either the house or the Senate. If it passes one then it goes to the other. If it passes both then it goes to the President to be signed into law.
E: technically there is an exception that bills for raising revenue have to originate in the house but that the Senate can propose or concur with amendments. But for all intents and purposes the vast majority of bills can originate in either body.
Is this just "AI porn bill" because that's the most common way of doing it these days? I should expect the product is what's being sanctioned and not the method.
…any intimate visual depiction of an identifiable individual created through the use of software, machine learning, artificial intelligence, or any other computer-generated or technological means….
The term ‘digital forgery’ means any intimate visual depiction of an identifiable individual created through the use of software, machine learning, artificial intelligence, or any other computer-generated or technological means, including by adapting, modifying, manipulating, or altering an authentic visual depiction, that, when viewed as a whole by a reasonable person, is indistinguishable from an authentic visual depiction of the individual.