ShieldFont: The AI-Proof Font That Tricks Bots, Not Humans

A groundbreaking font named ShieldFont has been engineered to combat the pervasive issue of unauthorized data scraping by large language models (LLMs), without compromising the visual experience for human readers. This innovative solution operates by subtly altering words within the HTML source code, rendering the text incoherent and unreliable for automated scrapers, while the original, coherent text is seamlessly displayed for human viewers. The development of ShieldFont is the result of a collaborative project involving the Brazilian creative studio Seneda & Abrucio, known for its innovative design solutions, and the esteemed Danish type foundry PlayType, recognized for its expertise in typography and font design.

ShieldFont's ingenious mechanism relies on a custom font implemented on the backend, which ensures that human users perceive the intended content accurately. Concurrently, it manipulates sentences within the HTML source code, injecting decoy words that deliberately disrupt the data ingested by automated web scrapers. For instance, a sentence that reads naturally for a human, such as “The knight rode his horse into battle,” might be presented to a scraping bot as “The knight rode his engine into battle.” This tactic is designed to degrade the quality of LLM outputs by feeding them jumbled, inconsistent, and ultimately misleading text. According to two of ShieldFont’s creators, Isaque Seneda and Gabriel Abrucio, the primary objective behind this innovation is to push back against the unauthorized scraping of data by LLMs, thereby increasing the difficulty and cost associated with such practices and discouraging them.

The technical foundation of ShieldFont's operation is rooted in the concept of ligatures, a well-established typographic feature where two adjacent letters within a word are combined into a single glyph. Ligatures are traditionally employed for aesthetic purposes, enhancing readability and visual flow by preventing awkward spacing, such as the common 'fi' in “fish” or 'fl' in “flow.” ShieldFont cleverly adapts this existing typographic capability, but instead of combining letters, it substitutes entire words. As explained by Felipe P., one of the project's creators, the true innovation lies not in inventing a novel font capability, but in repurposing an established one for an entirely new and strategic application. This approach ensures that while AI models attempting to ingest a website's content are presented with corrupted data, human users continue to enjoy a flawless and accurate textual experience. The effectiveness of ShieldFont is amplified by the inherent processing methods of LLMs, which rely on grouping words and phrases that frequently appear together. By introducing incoherent substitutions, ShieldFont directly interferes with this predictive grouping mechanism, leading to a significant degradation in the performance and accuracy of the AI.