He Scraped All of Their Art for AI. Now He’s

What happened

Zhang is frustrated not only by the scraping but also by the confusion around what she and the Cara team can realistically do to shield artists from malicious actors, saying that some users have already deleted their portfolios and abandoned the site.

” Once shaken out of this theoretical mindset and moved to make amends, Heft joined Cara’s Discord server as a troubleshooter and has been identifying why a number of proposed fixes are unlikely to be successful. “He’s just helping us, you know, taking the time to explain” the structural weak spots that allow for scraping, Zhang says.

When a certain tool or design comes up in the conversation, Heft lays out how he can “just break through in like a few minutes, literally,” she says. Because Heft is “of the belief that no site can be made truly ‘unscrapable,’” as he puts it, he and Zhang are collaborating on a countermeasure that can make a difference after the fact.

The tool, Lantern, allows artists to create a “one-way fingerprint” for their images without storing them on the platform. ” Lantern is just getting off the ground, and it’s the kind of imperfect workaround inspired by a regulatory vacuum.

Why it matters

“We have done the right things within limits without making it horrible to use,” Zhang says of the app’s current safeguards, including some new temporary measures like login gates—which, she adds, aren’t really a solution to an ongoing, internet-wide problem.

And Zhang worries that people blaming her for the string of attacks may not understand that Cara has almost certainly been scraped before, like any other site; it cannot guarantee complete security. “If it makes them feel better, deleting your work and leaving Cara, I support that,” Zhang says of those leaving.

“But I don't want to give people the misconception that if they go somewhere else, they are safer, because they're not. ” Yet Zhang, who is quick to remind WIRED that she is not a tech founder except by accident, has a newfound ally in this battle: the person who kicked off this month’s scraping frenzy.

After a Cara user confronted him and eventually persuaded to delete his dataset, she got in touch to understand his motivations and confirm the deletion. “He felt very bad to see how hurt people were,” Zhang says. ” “Heft” is a student in North America with a background in software and an interest in digital preservation and archival projects.

” In direct conversations with artists, he gained a greater appreciation of how personal their work was to them and how much they valued their ownership of it. “In retrospect, not only deliberately targeting Cara but presenting it the way I did in the post was cruel and thoughtless,” Heft says.

” He notes that while he was initially curious as to whether the data could be successfully used in AI training, he never believed it would, since 12 million images “is not a lot to train an image model” and “most commercial AI labs train from large-scale web scrapes” available from open-data organizations like LAION.

What to watch

But Zhang and Heft are aiming to give artists a little more control and a sense of awareness about how content gets hoovered up across the web, regardless of whether a website forbids this in its terms of service, as Cara does. Plus, since it’s open-source, anybody can contribute to the code.

Zhang’s biggest hope for the fallout from the scraping spree is that it might draw the attention of policymakers looking beyond questions of copyright, which is just one piece of a complicated puzzle. “Getting attacked by AI is just going to become so, so commonplace,” she says. And most of the time, the attacker won’t say sorry—let alone try to set things right.