A middle finger to AI scrapers

When generative AI image tools first arrived, they felt like magic. But to millions of visual artists whose work was scraped without permission or compensation, it felt more like corporate theft. Now, a startup is handing artists a digital crowbar to fight back.

Spawning, the team behind Have I Been Trained, has officially launched Kudurru, a new tool designed to block and sabotage AI scrapers in real time. The name comes from ancient Mesopotamian stone monuments used to mark land boundaries and property ownership.

Honeypots and middle fingers

To understand how Kudurru works, look at datasets like LAION-5B. These datasets do not actually store the images; they simply list billions of image URLs. When an AI company trains a model, it has to scrape and download those files directly from host servers across the web.

Spawning built a defense network of more than 1,000 honeypot websites hosting images cataloged in LAION-5B. When scrapers hit these sites, Spawning captures their IP addresses in real time. During internal testing back in July, Spawning says it successfully disrupted everyone downloading the LAION-5B dataset for roughly two hours. Currently, China leads all regions in total scraping volume.

Artists using Kudurru have two choices. They can flatly block the blacklisted IPs, or they can poison the data by serving unexpected content back to the scraper. Users can select what images to feed scrapers, and Spawning is not shy about its favorite option: sending back endless photos of a middle finger.

Over time, sustained poisoning can degrade prompt associations. If an illustrator poisons every scrape request, a model trained on that data might eventually associate their signature style with an offensive gesture.

Kudurru is launching as a free WordPress beta plugin, with plans to expand to audio and video. While tools like the University of Chicago's Glaze alter the image pixels to confuse models, Kudurru operates directly at the network layer.

Why it matters

For independent artists and photographers, Kudurru offers something long overdue: an accessible, zero-cost way to defend their personal portfolios. Enterprise-grade bot management tools exist, but they are far too expensive for solo creators. Kudurru levels the playing field just enough to give individuals a weapon.

It also shifts the dynamic from passive defense to active sabotage. AI companies have long treated the open web as free raw material. By poisoning training pipelines at the source, Kudurru threatens to spike data cleaning and filtering costs for model builders, forcing them to reconsider their harvesting tactics.

We think Kudurru is a brilliant tactical maneuver, but it remains a stopgap rather than a cure. AI firms can switch proxies and scrape through new infrastructure, turning this into an endless game of cat-and-mouse. As artist advocates rightly point out, creators should not have to carry the burden of policing multi-billion-dollar labs. Until legal frameworks mandate strict opt-in training standards, Kudurru is a sharp, necessary speed bump.