ShieldFont turns webpages into gibberish for AI scrapers
Designers Isaque Seneda and Gabriel Abrucio have built a font called ShieldFont that is meant to make a webpage readable for a human visitor while quietly serving AI scrapers a scrambled, nonsensical version of the same text. The font works through ligatures, the long-standing typographic feature that font engines normally use to redraw certain letter pairs as a single, more readable glyph when they sit next to each other. ShieldFont repurposes that mechanism: instead of merging two letters, its ligatures swap out entire words for different ones. That substitution happens only at the moment the font engine renders the page on screen. A scraper that just downloads the page's plaintext HTML source, rather than rendering it visually, ends up reading the altered version, the one a human visitor never sees because their browser renders the substituted text as ordinary, correct-looking words.
Seneda and Abrucio describe the goal, in a recent white paper, as giving web publishers "a practical opt-out from unauthorized AI training and [to] disrupt what is collected when that choice is ignored." The piece frames ShieldFont as the latest entry in a broader pattern of lawsuits and technical fixes that publishers have turned to in response to AI companies scraping large swathes of the public web for training data.
Choosing what to substitute is itself a design problem the piece lays out: swapping common words for plain synonyms or antonyms would be too easy for a capable scraper to detect and reverse, while replacing words with unrelated gibberish risks being caught, and potentially undone, by a scraping filter built to spot exactly that pattern. The piece does not explain how ShieldFont resolves that tradeoff, and it does not cite test data on how well the font holds up against real scrapers, name any AI companies or scraping tools it targets, give a release or availability date beyond calling the white paper "recent," or mention any publisher that has adopted it.
Key facts
- ShieldFont, created by designers Isaque Seneda and Gabriel Abrucio, shows human readers a normal webpage while AI scrapers reading the raw HTML source see a subtly altered, nonsensical version of the same text.
- The font repurposes ligatures, normally used to redraw adjacent letter pairs for readability, to swap out entire words instead; the substitution happens only when the font engine renders the page on screen.
- The designers describe the goal, in a white paper, as giving publishers "a practical opt-out from unauthorized AI training" and disrupting what gets collected when that choice is ignored.
- Choosing replacement words is a balancing act: plain synonyms or antonyms would be too easy for a scraper to reverse, while unrelated gibberish risks detection and circumvention by a scraping filter.
- The source gives no test data on real-world effectiveness, names no target companies or scrapers, and states no release date or publisher adoption.
Why it matters
AI companies scraping the public web for training data has already pushed publishers toward lawsuits and technical fixes aimed at stopping the practice, and ShieldFont is a new kind of response to that pattern. Rather than blocking scrapers outright, it lets the page load normally for everyone but quietly corrupts what a scraper reading raw HTML actually collects, without changing anything a human visitor sees. That is a different approach from access controls: it does not stop scraping from happening, it aims to make what gets scraped worthless as training data.
Who it affects
The designers frame ShieldFont as something for web publishers who want a way to opt out of having their content used for AI training without their permission. On the other side, it targets AI companies' scrapers: a scraper reading a page set in ShieldFont would collect the altered, nonsensical text rather than the real content. The piece does not name any specific AI companies, products or scraping tools as targets.
How to use it
The piece describes ShieldFont as a font that publishers would apply to a webpage so that AI scrapers reading the underlying HTML get the altered text while human visitors see the real content rendered correctly. It gives no detail on how a publisher would actually install or deploy the font, no pricing or licensing terms, and no release or general availability date beyond describing the white paper introducing it as "recent."
How solid is it
The piece lays out a real design tradeoff rather than presenting ShieldFont as a solved problem. Swapping common words for straightforward synonyms or antonyms would be simple for a capable scraper to detect and reverse; swapping them for unrelated gibberish risks being caught, and potentially undone, by a scraping filter built to spot exactly that kind of pattern. The article does not explain how ShieldFont's actual word substitutions are chosen to sit between those two failure modes, and it cites no test results or data showing how the font performs against real AI scrapers.
Risks and caveats
The scraping-filter risk the designers themselves acknowledge, that unusual substitutions could be flagged and reversed by a filter built for the purpose, is the central open question here. The source cites no evidence that ShieldFont has been tested against real scrapers, no publisher has been named as using it, and no timeline for release or wider availability is given. Whether the technique holds up once scraping tools adapt to it is untested by anything reported in the piece.
“a practical opt-out from unauthorized AI training and [to] disrupt what is collected when that choice is ignored.”
— Isaque Seneda and Gabriel Abrucio, ShieldFont designers