He Scraped Cara's Art for AI — Now He's Helping Build a Tool to Protect Artists
The artist platform Cara suffered three major data scrapes in August, but the fallout from the first led its perpetrator to help build a new open-source protection tool.
Since early 2023, photographer Jingna Zhang and a small volunteer team have run Cara, an image-sharing platform that has drawn around 1.5 million artists seeking an alternative to sites like Instagram, where content can be used to train AI without explicit consent. Cara filters AI-generated images and offers protective features, but fully preventing data scraping remains nearly impossible.
In August, Cara faced three major scraping incidents. On August 13, a Reddit user posted a 12-terabyte archive containing 12 million works — essentially Cara's entire public library — claiming the process cost less than $10. Cara's team learned of it through users who tagged them.
Zhang says the attack felt targeted and hurtful, noting that laws haven't caught up with such data harvesting, allowing scrapers to often claim their actions are technically legal. She is separately involved in two class-action lawsuits against companies including Stability AI, Midjourney, and Google over AI training on copyrighted artwork.
Two more incidents followed. A second scraper uploaded roughly 8.5 million links and metadata to Hugging Face; the platform agreed to remove personal metadata but said it could not remove the links themselves since no artwork copies were hosted there. A third scraper, on August 22, published 123,000 images along with personal user information on Academic Torrents. In response, Zhang launched a GoFundMe for legal fees with a $120,000 goal, raising more than $100,000 so far.
Unexpectedly, the person behind the first scrape, who goes by "Heft," came to regret his actions after Zhang confronted him, and he deleted the dataset. He said he hadn't anticipated how much distress the scrape would cause artists and called his framing of the Reddit post "cruel and thoughtless." Heft has since joined Cara's Discord as a troubleshooter, helping identify structural weaknesses that make scraping possible. Together with Zhang, he is now developing Lantern, an open-source tool that lets artists create a "one-way fingerprint" of their images without storing them on the platform. Lantern scans new public AI datasets and notifies artists if their work turns up, so they can pursue removal or takedown requests.
