AI Pulse by Inblix

ShieldFont feeds AI scrapers gibberish while humans see the real page

Ars Technica AI · Aug 12, 2026 · 2 min read · Read original article →

Curated by the Inblix editorial team


Featured image for article: ShieldFont feeds AI scrapers gibberish while humans see the real page

Two designers have built a font that exploits the gap between what a browser renders and what a scraper downloads. Isaque Seneda and Gabriel Abrucio call it ShieldFont, and the premise is almost too clever: use ligatures — the typographic trick that swaps ugly letter pairs for polished glyphs — to quietly replace entire words in the underlying HTML. Humans see the real article. Bots scraping the raw source get a corrupted version where, as their white paper puts it, a horse might become a potato.

The mechanics are worth understanding. Ligatures normally exist to make text prettier, merging ‘fi’ or ‘fl’ into a single smooth character. ShieldFont weaponizes that feature. The substitution happens only at the render layer, so any scraper that pulls plaintext HTML before a browser draws it gets nonsense. End users never notice a thing. It’s a fundamentally different approach from robots.txt blocks or paywalls, which scrapers can simply ignore — ShieldFont doesn’t ask for permission, it just makes the stolen data worthless.

Seneda and Abrucio are refreshingly clear-eyed about the arms race they’re entering. Not all word swaps are equally effective. Swap common words for obvious synonyms and a smart scraper will reverse the substitution in minutes. Replace them with total gibberish and a filter might flag the text as corrupted and bypass it. The designers describe aiming for a middle ground: substitutions that look plausible enough to evade detection but degrade the training value enough to make the scraped data useless.

This lands in a legal landscape where OpenAI, Google, and others are already fighting lawsuits over training data, and publishers are deploying technical countermeasures with mixed results. ShieldFont is a different bet — that the best defense isn’t blocking scrapers but poisoning what they take. Whether scrapers adapt by rendering pages like a browser does, which would neutralize the trick entirely, is the obvious next question. For now, it’s a cheap, clever middle finger aimed at anyone hoovering up the open web.

💡 Key Takeaways

  1. ShieldFont uses ligature substitutions to show readers real text while serving scrapers a subtly corrupted version in the raw HTML.
  2. The designers deliberately avoid both obvious synonyms and pure gibberish to make substitutions harder to reverse or detect.
  3. Unlike robots.txt or paywalls, ShieldFont doesn't block scrapers — it degrades the value of whatever they manage to collect.
  4. The font's weakness is scrapers that render pages like a browser, which would bypass the substitution layer completely.

Keep reading: See related articles below for more coverage on this topic.

Get smarter about AI

The sharpest AI news, curated daily. Delivered free to your inbox.

Learn more

Glossary terms

← Back to all articles