May 24, 2026
Fonts gone wild
Noroboto: Lying Fonts and Mitigation in Rust
A sneaky font fooled document-reading AI, and the comments instantly went feral
TLDR: A legal tech team found that a malicious font in a PDF could make software misread text while humans see something normal, raising scary questions about AI in law. Commenters were split between “this is a real risk” and “this is just fraud with extra steps,” with plenty of jokes about screenshots and alphabet-swapping fonts.
A developer thought they’d found a bug while rebuilding a document-reading tool in Rust, a programming language loved for safety and reliability. Plot twist: the bug wasn’t just in their code. A weird font trick inside a PDF could make text look normal to humans while feeding the wrong meaning—or missing letters entirely—to software. In a world where law firms increasingly let artificial intelligence, or AI, chew through contracts and court filings, that’s the kind of discovery that makes everyone sit up straight.
But the real fireworks were in the comments, where the crowd split into two camps: “terrifying exploit” versus “come on, this would never survive in real court.” One skeptic basically laughed that if you’re going this far, you may as well paste a screenshot into Word and call it a day. Others piled on with even simpler chaos ideas: why stop at one sneaky letter when you could make a font where every A is secretly a Z? Another mini-drama broke out over ligatures—those little font combos that merge letters—because several readers argued the proposed fix might be easy to dodge.
And then came the legal pearl-clutching. Some commenters said any lawyer trying this would be flirting with plain old fraud, not some genius courtroom hack. Still, beneath the memes and eye-rolls, there was grudging respect for one thing: using Rust made it easier to trace the problem and build defenses. So yes, the internet did what it does best—joke first, doom-post second, and accidentally surface the real issue in between.
Key Points
- •Tritium is replacing PDFium with the Rust-based hayro crate to reduce platform friction and better support WASM-based web deployments.
- •During that migration, the author discovered a text extraction issue in which copied PDF text gained an extra space and lost characters.
- •The investigation found a double-t glyph represented by a non-Unicode "tt" value that hayro treated as a non-breaking space, while PDFium also appeared not to handle it correctly.
- •The article argues that legal-tech pipelines are complex combinations of open-source and proprietary document tools that may contain exploitable implementation imperfections.
- •The post introduces noroboto.ttf as a conceptual malicious embedded font designed to obfuscate or alter glyph Unicode mappings and thereby confuse AI agents in legal document workflows.