Cara faces repeated scraping attacks, and launches Lantern alerts

In short: Cara, an art portfolio app built to avoid AI training use, was repeatedly scraped in August, and its team is now working on a new tool that alerts artists when their work shows up in public AI datasets.

What's going on

Cara is an image sharing and portfolio platform started by photographer Jingna Zhang and volunteers. It has drawn about 1.5 million artists who want to show their work without making it easy for AI companies to use it for training.

In August, Cara was hit by several large scraping incidents. Scraping is when someone uses software to copy lots of data from a website quickly, like taking every page from a library at once instead of checking out one book.

One person posted that they collected a 12 terabyte archive of about 12 million works from Cara and shared it on Reddit, saying it cost under $10 to do. Zhang said the scrapes also drove up Cara’s server bills. A second incident involved about 8.5 million links and other details like usernames and tags being uploaded to Hugging Face, a site where developers share AI related files. A third scrape shared 123,000 images and text posts, including some personal information, on Academic Torrents.

Zhang launched a GoFundMe with a $120,000 goal for legal fees, and said it had raised over $100,000 as of the report.

What to watch

In a twist, the first scraper, a student who goes by “Heft,” apologized and agreed to help. He and Zhang are now collaborating on Lantern, an open source tool that makes a one way fingerprint of an image (like a unique ID card number) and scans new public AI image datasets. If it finds a match, Lantern can notify the artist and link to the dataset so they can ask for removal or file a takedown request.

Source: Wired