Brazilian creative studio Seneda & Abrucio and Danish type foundry PlayType have released ShieldFont, a typeface designed to confuse AI scrapers.
The tool represents a new front in the conflict between content creators and AI companies. By making text difficult for bots to read accurately, creators can protect their intellectual property from being used to train large language models (LLMs) without permission.
Launched on July 22, 2026, the typeface functions by swapping words behind the scenes [3]. While a human reader sees a standard sentence, the underlying data processed by an AI agent is altered. This discrepancy is intended to poison the datasets that bots collect from the web [1], [2].
Developers of the font said the goal is to create a barrier against the automated scraping of digital text. The mechanism ensures that the data ingested by LLM bots is inaccurate or nonsensical, effectively degrading the quality of the training set [1], [2].
According to reports, "ShieldFont was designed to poison the well of data scraped by LLM bots" [2]. The collaboration between the Brazilian and Danish teams focused on a visual experience that remains seamless for people, while remaining opaque to machine learning algorithms [1].
This approach differs from traditional robots.txt files, which request that bots ignore certain pages. ShieldFont instead allows the bot to enter but provides it with corrupted information, a strategy known as data poisoning [1], [2].
“ShieldFont was designed to poison the well of data scraped by LLM bots.”
The emergence of AI-proof typography signals a shift from passive opt-out requests to active technical countermeasures. As AI companies rely on massive amounts of web-scraped data to improve model accuracy, tools like ShieldFont create a 'cat-and-mouse' game where designers attempt to outpace the pattern-recognition capabilities of machine learning.
