Anti-AI art platform Cara was scraped in full for under ten dollars

Cara exists because artists wanted one place online that would not feed their work to AI. Someone pulled roughly 12 million images off it for under ten dollars and called it a fun project.

What Cara was built to prevent

Photographer Jingna Zhang launched Cara in 2023 as a portfolio platform for artists who were leaving larger networks over AI training. It ran anti-bot and anti-scraping measures. Its robots.txt told crawlers not to take work for AI. Uploads were tagged NoAI automatically. It had opted out of federated services.

On August 14, 2026 a Reddit user posted: “I scraped all of Cara, an anti-AI art platform! Was a fun project, happy to answer any questions here.” PetaPixel put the haul at just over 12 terabytes, obtained at a cost of under ten dollars. AsiaOne reported roughly 12 million images belonging to more than a million users. The post was later deleted and the account was banned from the subreddit.

Zhang’s reaction on X was blunt: “What the hell must I do? Artists just want a place online where platforms don’t feed their work to AI.” To the argument that the site should simply have been harder to scrape, she replied: “A door not being bank vault-strong isn’t an invitation to violate.”

Opt-out signals are requests, not locks

The episode exposes the uncomfortable center of every opt-out mechanism in the AI training debate. robots.txt directives, NoAI tags and platform terms of service are all requests. They are readable by anyone who intends to honor them and invisible to anyone who does not. Publicly visible images are, by definition, retrievable, and the cost of retrieving them at scale is now less than lunch.

That is not an argument that Cara did anything wrong. It is an argument about what those measures can and cannot deliver. Copyright litigation and data provenance standards are moving, slowly, but consent signals on their own do not stop a determined scraper with ten dollars to spend.

What artists should take from it

The practical lesson is about expectations. A platform that promises to keep AI out can honestly promise to refuse cooperation, to signal non-consent in every machine-readable way, and to make scraping inconvenient. It cannot promise that publicly displayed work will never be copied. Artists building a refuge deserve to know that distinction before they move in. A rule that costs ten dollars to break is not protection; it is etiquette, and etiquette only binds the people who already agreed to it.