Updated
Updated · Forbes · Aug 23
Cara Suffers 3rd Scraping Attack in 10 Days as 123,000 Images Hit Academic Torrents
Updated
Updated · Forbes · Aug 23

Cara Suffers 3rd Scraping Attack in 10 Days as 123,000 Images Hit Academic Torrents

2 articles · Updated · Forbes · Aug 23

Summary

  • Academic Torrents hosted a new Cara dataset on Aug. 22, marking the artist network’s third scraping attack in 10 days and exposing copyrighted images, metadata, text posts and user bios.
  • 123,000 images were taken in the latest scrape, while founder Jingna Zhang said the repeated attacks now appear targeted at artists who oppose generative AI and rely on Cara’s consent-based rules.
  • Nearly 12 million Cara images were scraped in the first known attack and posted to Reddit, then a second dataset appeared on Hugging Face, where the platform said it hosted only URLs and metadata, not the artworks themselves.
  • 1.5 million users have joined Cara since 2022 as a refuge from AI training practices, but Zhang said the scrapes are adding thousands of dollars in server costs and forcing a GoFundMe push for legal defense.

Insights

If open-research platforms distribute stolen art, who is truly liable for the unauthorized training of generative AI models?
Can an anti-AI sanctuary survive when scrapers turn its defenses into a target for massive data heists?