This Font Looks Normal to Humans, Nonsense to AI Scrapers

A human reader looking at a webpage using ShieldFont sees "The knight rode his horse into battle." The HTML the same page serves to an AI scraper reads "The knight rode his engine into battle."

That gap is the trick. Designers Isaque Seneda and Gabriel Abrucio built the open-source typeface to swap words in page source code with random substitutes, tuned to look like ordinary text to a naive crawler but chaotic enough to corrupt a training corpus. The project's white paper reports ShieldFont replaces about 24.4 percent of all words and 42.8 percent of content words on a typical page. The swaps ride on ligatures, a typeface feature most publishers and CMS pipelines already support, so deployment is cheap. The code is on GitHub.

The creators call ShieldFont a "consent layer," not a blocker. From the white paper: "Nothing currently makes it costly to ignore a publisher's wishes. This paper explores a different approach: making the text itself polluted, harder, and more expensive to collect without permission." They note an "advanced scraper" can still detect and reverse the swaps, framing the project as a prototype rather than a fix.

Ars Technica, The Register, and Help Net Security have covered the project. What remains open is whether scrapers will adapt, and whether other publishers will adopt a font as a policy surface.