ShieldFont uses a font to confuse AI web scrapers

In short: ShieldFont is a new font that keeps web pages readable for people but turns the text that many AI scrapers collect into nonsense.

What happened

Two designers, Isaque Seneda and Gabriel Abrucio, introduced ShieldFont, a font meant to help website owners opt out of having their writing collected for AI training without permission. Many AI companies use “scrapers,” which are bots that automatically copy huge amounts of public web pages to use as training data.

ShieldFont works by showing one set of words to people and a different set of words to bots. It does this using “ligatures,” which are a normal font feature where certain letter combinations get drawn as a single shape. In ShieldFont, that same idea is used to swap whole words only when the page is drawn on your screen. The page can look normal to a reader, but the raw HTML (the page’s underlying text and structure, like a script behind a play) contains different words that a scraper might copy.

Instead of swapping words for simple opposites or random gibberish, ShieldFont replaces them with other real words that fit the grammar but change the meaning, like swapping “horse” with “potato.” The creators say their current list covers nearly 12,000 common words and replaces about a quarter of the words on an average page.

Why it matters

This could make large-scale scraping more expensive and less useful, since scrapers may either reject the page as low quality or accept text that sounds fine but is not true. But it has tradeoffs. Search engines, screen readers for people with vision loss, translation tools, and even copy and paste can be affected because they read the underlying text, not what the font draws.

Source: Arstechnica