Cara keeps getting scraped. The artist social network has prohibited AI access under its service terms, yet data miners have copied its work three times in ten days, publishing copyrighted images and metadata on external platforms.
Founded by Jingna Zhang, Cara has attracted about 1.5 million artists by filtering out AI images and offering protective features including Glaze. That makes the service a clear target for people collecting material for AI systems — and a particularly frustrating one for artists who joined to keep their work away from those systems.
The sequence began on August 13, 2026, when a 12-terabyte archive containing 12 million works from Cara appeared on Reddit. The Reddit user responsible was MandarinDawnPoppy994, who described the effort with four words: “It was a fun project.”
That archive was only the first breach in the sequence. The second scrape pulled about 8.5 million links, usernames, titles, and tags from Cara, then uploaded the material to Hugging Face, an AI developer platform.
The third scrape collected 123,000 images from Cara, along with text posts and user bios, before sharing them on Academic Torrents. The result was not just a pile of images: it included metadata and personal material tied to the people who posted on Cara.
Three Scrapes, One Unpleasant Pattern
The second scrape was reported on August 22, 2026. A third disclosure followed on Thursday, shortly after August 22, adding another archive of Cara content to the list.
Zhang said Cara learned about the first scrape through its own community: “We actually found out about it through our users tagging us.” That detail captures the platform’s position. Cara can prohibit AI access in its terms, but the people using the service often discover violations before the service does.
The scale changed with each incident. The first archive contained 12 million works and consumed 12 terabytes; the second recorded 8.5 million links and pieces of metadata; the third contained 123,000 images plus text posts and user bios.
Those numbers describe three different forms of exposure. One copied a huge body of artwork, another extracted the connective data around artists and posts, and the third combined images with written material and bios. The internet remains very good at turning boundaries into suggestions.
Zhang described the impact in personal terms: “I just feel it’s targeted and very hurtful.” For Cara’s users, the issue is not only whether an image enters an AI dataset. Their usernames, titles, tags, text posts, and bios have also appeared beyond the service’s intended boundaries.
From Scraper to Unlikely Helper
MandarinDawnPoppy994 later decided to help Cara after seeing the response to the scrape. Zhang said, “He felt very bad to see how hurt people were. So he decided to help us.” The quote creates an unusual turn in a story otherwise built around extraction and resistance.
The facts do not make that turn tidy. The person responsible for the first scrape remains connected to the archive that triggered the backlash, while Cara still faces the publication of copied material across Reddit, Hugging Face, and Academic Torrents.
Zhang launched a GoFundMe with a goal of $120,000 to cover legal fees. The campaign has raised more than $100,000, showing that Cara’s large artist community has responded with money as well as warnings and tags.
Cara’s model depends on a simple promise: artists can share work on a platform that excludes AI images and blocks AI access through its service terms. Three major scrapes in ten days have tested that promise from every direction, exposing the gap between what a platform forbids and what determined data miners can publish elsewhere.
The archives also make the dispute concrete. This is not an argument about an abstract future dataset; it involves 12 million works, 8.5 million links and metadata records, and 123,000 images accompanied by text posts and user bios. Cara now has a legal fund, a community watching for violations, and an unexpected helper who once called the first scrape “a fun project.”
Based on
