From Recorded Future News and PRX, this is Click Here. There used to be a word for this when someone took your work and passed it off as their own. We called it forgery or plagiarism. It was clear and, at least in theory, settled. But now, some of the biggest companies in the world are doing something that looks a lot like it. And they're calling it innovation. The new, new thing. Generative AI is trained on the internet. Which is another way of saying it's trained on us. On millions of pieces of human creativity. Paintings, songs, voices. all scraped, absorbed, and reassembled into something that feels new, or at least passes as new. The top result isn't even my art. It's art made from my art. If a program could do that, you know, what's the point? So what do you call it when the line between original and imitation just disappears? From Recorded Future News and PRX, this is Click Here, a podcast about how technology is changing everything. This week, Karen Duffin looks at what happens when creativity itself becomes raw material. It's forcing artists to ask themselves what it means to create when their work is training a machine. And what happens to that machine when humans try to break it? You can fight back with the same technology that they're exploiting you with. That's after the break. Stay with us. Looking for more of the cybersecurity and intelligence coverage you get on Click Here? Then check out our sister publication, The Record, from Recorded Future News. You'll get breaking cyber news from reporters in New York, Washington, London, and Kiev, among others. And you'll see for yourself why it attracts hundreds of thousands of page views every month. Just go to therecord.media. I'm Karen Duffin, and this is Click Here. If you are an artist, you know how long it can take and how hard it can be to develop a style that's all your own. A painter with a particular palette and with a signature sound. And it took time, but artist Kelly McKernan did just that. I work in watercolor and acrylic wash, and my work is vibrant and ethereal. Kelly is a painter in Nashville and goes by the pronouns they, them. Their work blends human and abstract images, capturing introspective and almost spiritual themes. I've done some work for Dark Horse Comics. I've also worked with Evanescence, the band. I've done some comics as well. A few years ago, they started getting tagged in images on Instagram and X. Not too unusual. They have tens of thousands of followers. But when they clicked through and looked at the images, it was confusing. When that started to happen, I didn't really understand what that was about, especially because it didn't look like my work, but it was using my name. It didn't take long until they found their answer. I discovered that more than 50 of my paintings had been scraped to use as training data for AI image generators. Kelly's art had been vacuumed up along with billions of other images for anyone to play with. Users could open this image generator and ask it to make AI paintings in the style of Kelly McKernan. And then these programs will generate images, several images, created with my style based on these 50 or so paintings that were scraped. Apparently, a lot of people were doing this because now when Kelly Googled their name. The top result isn't even my art. It's art made from my art. And that's a big part of the problem is if somebody can just generate something that looks pretty and good enough, they're not going to hire me. I felt violated. I was upset to see this happening. I was even more upset that it had happened without my consent and that not a single one of these people who were using my name as a style prompt had even thought to ask me how I felt about this. Kelly had done the nearly impossible for 15 years by then. They made a living mostly supporting themselves as an artist along with some teaching But after this database came out We started seeing the small projects fall away The three or four projects in a month might pay my rent turned into one to zero, like complete crickets. But the impact for Kelly wasn't just financial. It was almost existential in a creative sense. their style, their voice. It used to feel uniquely theirs. I really questioned the point of not just my work, but of creating art at all. If a program could do that, you know, what's the point? It was pretty dark at the end of 2022, really just kind of, yeah, what's the point of everything I'm going to make is going to be scraped and used to essentially displace me. Almost immediately, Kelly started asking what you might be asking yourself right now. How is this legal? And that is a question Kelly and other artists have recently started asking the courts to answer. An unprecedented class action lawsuit along with two other visual artists against AI companies. A group of prominent authors is suing the company behind ChatGPT. Says he and thousands of other writers have been ripped off. In 2023, Kelly became a lead plaintiff in a class action lawsuit, along with other artists. That same year, a group of authors sued OpenAI on similar grounds, including journalist Ta-Nehisi Coates and comedian Sarah Silverman. Both suits alleged copyright infringement. There are people who are saying because all of this comes from stolen, pirated art, well, AI companies actually, they should pay because they were trained on people's works, right? Matt Blaschik is a lawyer who studied technology and copyright. He's at the University of Michigan. And he says the arguments go like this. On one side, Kelly and other artists argue that these AI systems couldn't exist without their work. And on the other side, AI companies say that what they're doing is legal. They point to a corner of copyright law called fair use. Essentially, it says you can use other people's art as long as you're referencing, remixing, or building on the source material. Like when a singer like Weird Al Yankovic writes a song like Eat It. Even though it mirrors Michael Jackson's song Beat It so closely. Something like that is generally considered fair use. It's based on something original, but so thoroughly remixed that in a sense, it's something new. And that is what AI companies are banking on in their defense. But there's also another argument here, baked right into the essence of copyright law. Something much trickier to discern. Copyright access to expressions, not ideas. So some people would claim. Does Copyright 101? In other words, copyright protects the exact thing you made, but not the style behind it. Which leaves the courts to answer two big questions. First of all, did they actually copy your expression? Did they take what's yours and then implement it in the copy? And the other question is whether the two works are substantially similar, right? Is the new work, does it look like the one you created? Did the system take something that's recognizably yours? and is what it produces not just close, but too close? And these are hard questions for any court to answer. Harder still when the law they're leaning on hasn't really been updated since 1976. That's when Congress codified the idea of fair use. 50 years ago, back when phones had chords and music came on eight tracks. The internet wasn't even a thing yet, AI less than a glimmer. And now that same law is being used by courts to make sense of a world it never anticipated. Because as companies like Suno and the non-profit Lion change how we make and consume art, it's forcing courts to consider questions that lawmakers in 1976 could never have imagined. Like who gets credit when a machine learns from your work If your work is one of one billion that went into a data set which then results in the creation of a new painting a new image a new song what kind of credit would be even sensible to speak of? And what kind of compensation would be appropriate? So far, courts haven't given a clear answer. In early 2025, a judge ruled against that group of authors who made claims similar to Kelly's, but parts of their case have yet to be decided. Kelly's case is still in process. On the other hand, Anthropic agreed to pay a $1.5 billion settlement after a judge found the company trained its AI on millions of pirated works. It's reportedly the largest copyright payout in history. And with such mixed legal messages, it leaves artists in a difficult spot. Wait for answers or find ways to fight back. There is a potential that this actually poisons the overall algorithm. As you give it something to train, and as it learns from that, then it actually gets worse at training. That's when we come back. Stay with us. The Wired Newsroom is known for award-winning reporting on how technology shapes our world. On Wired's Uncanny Valley, we take that curiosity even further. Each week, journalists from Wired break down the biggest stories in tech while speaking directly with the people building, challenging, and reshaping the future. Is the AI boom sustainable? How do you protect your privacy in an age of constant surveillance? Uncanny Valley tackles the questions driving today's tech debates and lighting up your group chats. Listen to new episodes every Thursday, wherever you get your podcasts. It is called internet. I use the World Wide Web information superhighway. Cybersecurity. Why do things go viral? Click here. Ben Jordan is a musician known as the Flashbulb. He's also a popular science YouTuber. And his fight against AI is a little different. A little more, well, punk rock. Because instead of going to court, Ben started experimenting. A few years back, he figured out how to stop microphones from recording his concerts without his permission. The trick is something called adversarial noise. And so I was like, okay, what if we just have a really loud ultrasonic tone that a phone's microphone or a Google Home or an Amazon Echo could pick up but a human can't hear? Think of it like a dog whistle. Humans can't hear it, but machines can. So Ben started bringing ultrasonic sensors to his shows and turning them on while he performed. If you're pointing a phone at me and listening to me playing the guitar or something, it will just jam the microphone and it'll sound like a distorted mess and you won't be able to hear anything. In other words, to a person, the music sounded great. But to a machine, it's unusable. And that got Ben thinking. If he could scramble a recording, could he take it one step further and scramble the data AI systems learn from? Is it possible to do that in a way where we actually encode a file with noise that humans can't hear? However, AI algorithms that are training on the data would hear this. So he went back to his computer and started tinkering with sound files, adding layers, hiding signals, building what are essentially booby traps for AI. There's really sinister ways of doing this. It's just that AI hears music very differently than we do. And somewhere between those two areas, there's room for something to fit that actually obfuscates the data for AI while keeping it completely sounding untouched to human ears. Here's what he means. Humans hear music as sound, but AI reads it as an image. Before AI systems can learn from a song, they have to translate that song into images, a kind of visual map with peaks and valleys, lines and blobs. So Ben didn't need to wreck the song. He just needed to tamper with the picture the machine is reading. And the result? To us, to humans, the music sounds untouched. To the AI, it's a mess. So here we go. We can upload my original song here. And here is Suno AI extension There a name for this approach Data poisoning The idea is simple. Feed the AI system bad information so it learns the wrong thing. Ben's creation is a tool he calls Poisonify. So the way that this works is it confuses it. So it thinks that the drums might be a piano and it thinks that the guitar might be a harmonica and it thinks the vocals might be a trumpet. And so it's just kind of useless for training. It'll sound wrong and it'll sound messy. Ben doesn't plan to release Poisonify to the public. For one, it takes a lot of energy to use. Encoding an entire album takes about two weeks of nonstop processing. But he does see it as a proof of concept. And he's taken what he's learned and collaborated with other researchers like the makers of Harmony Cloak, which works on a whole different level. What it mostly does is it takes harmony or melody and just makes it unrecognizable noise. It's no longer something you would want to listen to. While Poisonify is teaching AI the wrong instruments, Harmony Cloak is teaching it the wrong music altogether. So a song that sounds like this... ...gets scraped by AI and ends up sounding like this. The idea is that if you do it often enough, the entire data set will be corrupted, essentially poisoning the entire well. Ben envisions a future when musicians layer their work with these kinds of AI poison pills. As it learns from that, then it actually gets worse at training because now you have adversarial noise or a poison pill in the database. Like you can fight back with the same technology that they're exploiting you with. But, he says, it's not that these technical changes will stop AI companies. They aren't likely to give up without a fight. It's that he hopes they make it just uncomfortable enough that the companies decide it's easier just to work with artists than go around them. I do believe that this could launch an arms race, because if Suno has to cough up so much money to protect against technology like this, that they'll just start licensing music and paying musicians, then success. Like, the idea is not to destroy AI. The idea is just to get musicians fairly compensated. more so than anything. And in the meantime, he keeps making his music. The thing that AI can't do is take away the enjoyment of making music. That's why I'm not threatened by it one bit. It can't take away me sharing it with people that I love or even people who are in my community. That was Karen Duffin. This is Click Here. Click Here is a production of Recorded Future News and PRX. Today's show was written and produced by Megan Dietry, Sean Powers, Erica Gaida, Zach Hirsch, and Casey Georgie. It was edited by Karen Duffin and Sarah Covedo, and fact-checked by Darren Ancrum. Original music is by Ben Levingston, with additional music from Blue Dot Sessions. Our staff writer is Lucas Riley, our illustrator is Megan Goff, and our sound designers and engineers are Jake Cook and Jesse Neiswanger. Find us on X or Facebook at Click Your Show. Or leave us a voice message at 6615CHTALK. Sometimes we'll turn those moments into reporting, sometimes into a conversation, and sometimes into a future story you'll hear on this show. I'm Dena Templereston, and thanks for listening. Support for this program comes from Recorded Future. In cybersecurity, the biggest risk isn't what can be seen. It's what gets missed. Recorded Future analyzes billions of signals to help organizations stay ahead of threats. Recorded Future. Know what matters. Act first. Looking for more of the cybersecurity and intelligence coverage you get on Click Here? Then check out our sister publication, The Record, from Recorded Future News. You'll get breaking cyber news from reporters in New York, Washington, London, and Kiev, among others. And you'll see for yourself why it attracts hundreds of thousands of page views every month. Just go to therecord.media.
