21/08/2026
Until recently, publishers had two defenses against AI crawlers copying their content, and both rely on the crawler behaving itself. Robots.txt names which crawlers may enter, and compliance is voluntary, while Cloudflare blocks the known ones until a scraper changes its identity. ShieldFont, released in late July by independent studio S&A with Copenhagen foundry Playtype, gives up on the wall and hands the intruder a corrupted copy instead.
ShieldFont exploits a split between what you see and what a crawler copies from one page. You see the pixels a browser renders, while a mass scraper copies the raw HTML underneath, which means one page can show a reader one sentence and give a bot another.
Their own FAQ spells out the swaps, where "winners" turns into "avengers", a "toy gun" into a "sofa car", and a "Halloween contest" into an "Autumn campaign". A scraper records that a child found a "sofa car", while the font restores the original wording on screen for you.
Keeping those decoys grammatical is what does the damage, since obvious gibberish gets caught by the quality filters AI companies apply to scraped text while a fluent sentence stating the wrong fact passes straight through. Around 25% of words on a page are swapped, nearly half of the ones that hold the meaning, and 55.8% of shielded passages no longer made the same factual claim in controlled testing.
Its creators are candid about the limitations, since a screenshot passed through OCR defeats the entire method. The aim throughout has been cost. Reading raw HTML is cheap, and photographing millions of pages to extract the text costs a pretty penny.
Image credit: shieldfont.org