Designers and Computer Scientists Create New Decoy Fonts That Humans Can Read but A.I. Bots Cannot

To prevent the trawling and theft of human writing, font-makers are employing both old and new illusions to outsmart large language models

Christian Thorsberg | Daily Correspondent

September 11, 2026 12:08 p.m.

Share

Copy linkEmailSMSFacebookXRedditLinkedInBlueskyPrint

Add as preferred source

As the deployment of A.I. tools into search engines, text messaging and writing platforms becomes increasingly conspicuous and pervasive, some researchers are seeking new ways to insulate words written by humans from the reach of trawling bots and A.I. agents.

The camouflaging technique is a modern-day version of an optical illusion known as hybrid imaging, which was made famous in 2007 when researchers at MIT created a portrait in which both Marilyn Monroe and Albert Einstein were identifiable, depending on the distance of the viewer.

Whereas those images were created to better understand human perception, Decoy Font was inspired by a desire for privacy. The font features two layers of text: one foregrounded in clear outlines and a second dark and blurred in the background. It offers human writers the chance to create cheeky juxtapositional messages, hiding the blurred “Happy Human,” for example, behind a crisp “Sorry Robot.”

MixFont’s first attempt at an anti-A.I. typeface was called Ghost Font: a slightly dizzying display that placed moving words atop a television-static-like background that also constantly shifted. The company acknowledged that it wasn’t a traditional font, because the message was readable only  in a video format, but the design successfully demonstrated the limits of A.I. perception.

“I was inspired to make Ghost Font one day after texting with my mom, and I noticed that A.I. kept trying to summarize our messages, many times incorrectly,” Eric Lu, MixFont’s founder, tells Forbes’ Leslie Katz. “I was annoyed that it was being so intrusive in our private conversation, and I started brainstorming ways that I could potentially communicate without A.I. reading everything.”

Other designers have also experimented in recent months with tricking A.I. trawlers.

In July, a collaboration of studios including Seneda & Abrucio and PlayType released ShieldFont, a typeface that scrambles text for automated scrapers. The font replaces roughly half of all “content words”—nouns, verbs, adjectives and adverbs—with random other words in its source code, such that the writing taken by bots is nonsensical. More than 90 percent of the resultant gibberish is rejected by quality filters, Fast Company’s Hunter Schwarz reports. 

“A sentence is a sequence of decisions,” ShieldFont’s creators write in their white paper, explaining that a primary motivator for their work was to block the theft of human writing by A.I. models. “When I write, I am choosing this word rather than that one, putting this clause before that one, deciding what to leave out. The finished paragraph is a record of thinking that took effort to do. ... We are being asked to hand over the work, and to train our own replacement with it, and to do both without being told.”

ShieldFont: A Typeface that Poisons Unauthorized AI Training

Watch on

MixFont “started as an exploration of how A.I. and typography intersect, but it was also meant to raise awareness of the growing reach and capabilities of A.I.,” Lu tells Dezeen’s Rima Sabina Aouf. “I think the reaction to the font shows that there are many folks who feel strong A.I. fatigue and want a way to defend themselves against the presence of A.I. in their lives.”

Christian Thorsberg is an environmental writer and photographer from Chicago. His work, which often centers on freshwater issues, climate change and subsistence, has appeared in Circle of Blue, Sierra magazine, Discover magazine and Alaska Sporting Journal.