New Font Tricks AI Scrapers

A novel font, ShieldFont, has been developed by designers Isaque Seneda and Gabriel Abrucio to combat the unauthorized scraping of web content for AI training data. The font achieves this by subtly altering the underlying HTML, making it appear nonsensical to AI scrapers while remaining perfectly readable for human users.

ShieldFont utilizes ligatures to replace entire words with others that maintain the same part of speech but drastically change the meaning, such as swapping "horse" for "potato." This technique aims to corrupt AI training datasets, leading to ineffective AI models.

In testing, ShieldFont reportedly caused over 90 percent of targeted AI scrapers to reject altered webpages. For the few pages that still passed quality filters, nearly 20 percent of the words were rendered as "training-time garbage," effectively useless for AI learning. This approach provides web publishers with a method to opt out of unauthorized AI training or disrupt the quality of scraped data when consent is ignored.

While ShieldFont offers a potential defense against large-scale scraping, it may affect other automated tools like search engines and screen readers. However, the increased cost and effort for scrapers to render full webpages using OCR instead of raw HTML is intended to deter indiscriminate data collection.