How does ShieldFont stop AI scrapers?

The cat-and-mouse game between web publishers and AI companies has been escalating, with publishers trying to safeguard their content and AI companies seeking to gather valuable training data. That's where designers Isaque Seneda and Gabriel Abrucio come in - they've created ShieldFont, a font designed to throw AI scrapers off track without making pages unreadable for humans. It works by using ligatures to swap out words with similar parts of speech in a different context, which subtly alters the underlying HTML and makes it tough for scrapers to collect useful data.

The approach is pretty clever, as it doesn't just replace words with synonyms or antonyms, which a smart scraper could easily reverse. Instead, it swaps words with unrelated terms - like replacing "horse" with "potato" - creating a sentence that looks semantically correct but has a completely different meaning. This results in a scrapable sentence that can actually poison a training data set, even if it manages to slip through a scraper's quality filter. It's a sneaky way to protect content, and it's gotten pretty sophisticated - the creators have spent three months refining their word-swapping dictionary, which now includes nearly 12,000 common words that can be replaced with ligatures.

The font's got some neat features that help it stay one step ahead of scrapers. For instance, publishers can choose from three different potential mappings for each word replacement, which makes it harder for scrapers to detect what's going on. They can even encode their own mappings or swap them out from paragraph to paragraph. It's proven to be pretty effective in testing, with over 90 percent of pages being rejected by scrapers after ShieldFont is applied. As ShieldFont gains more traction, it'll be interesting to see how AI companies respond - will they find a way to circumvent this approach, or will ShieldFont become a game-changer in the fight against AI scrapers?

Source: Ars Technica