Cara Scraper Helps Build Lantern Tool to Protect Artists

Following a series of massive data scrapes on the artist portfolio platform Cara, an unlikely partnership has emerged to fight unauthorized AI training. Jingna Zhang, the founder of Cara, is collaborating with a software student known as Heft to develop Lantern, an open-source tool designed to help artists monitor where their work ends up. Heft was the very individual who initiated the scraping wave on August 13, extracting a 12-terabyte archive of 12 million works from Cara for less than $10.

The incident was the first of three major scrapes that targeted Cara, an app that has attracted 1.5 million artists seeking refuge from AI exploitation. Shortly after Heft's scrape, a second actor named CaptiveDreamer pulled 8.5 million links and metadata from Cara, uploading them to Hugging Face. On August 22, a third scraper harvested 123,000 images and user bios, publishing them on Academic Torrents. In response to the attacks, Zhang launched a GoFundMe campaign to cover legal fees, raising over $100,000 toward a $120,000 goal.

After deleting his dataset and apologizing, Heft joined Cara's Discord to help identify security vulnerabilities. Because Heft believes "no site can be made truly 'unscrapable,'" the duo focused on post-scrape detection. Their new tool, Lantern, generates a "one-way fingerprint" of an artist's image without storing the file itself. Lantern continuously scans newly released public AI training datasets. If a match is detected, the system alerts the artist and provides a direct link to the dataset so they can file a takedown request.

For creative practitioners and AI developers, Lantern introduces a decentralized layer of accountability in the data-sourcing pipeline. While it cannot physically block scrapers, it gives artists a systematic way to audit AI models and enforce their intellectual property rights. Since the project is open-source, developers worldwide can contribute to its code, potentially establishing a new standard for tracking data provenance in the machine learning era.