Scientists Develop 'Ghost Font' to Evade AI Detection

Scientists have developed a new font, dubbed 'Ghost Font,' designed to be readable by humans while baffling artificial intelligence systems. The innovation uses a combination of motion, video, noise, and decoy messages to obscure the intended text from AI detection.

Eric Lu, the developer, described Ghost Font as an experiment in graphical communication that moves beyond traditional font files. While the letters are perceptible to the human eye through their movement, AI models have struggled to decipher them even when presented with videos of the font. Lu noted that even advanced AI models like Fable and GPT Sol 5.6 Ultra have difficulty decoding messages written in Ghost Font.

The concept draws inspiration from ZXX, a typeface created in 2013 by designer Sang Mun, which aimed to be unreadable by optical character recognition software. However, Lu noted that modern AI can now easily read ZXX. To counter contemporary AI capabilities, Ghost Font relies on motion. When a message is typed, the letters appear through the movement of dots, but when the video is paused, the static dots merge, making the message invisible in a single frame and resistant to screenshots.

An additional security feature includes a decoy message, designed to mislead AI agents by presenting them with an apparent hidden message that is not the true content. This layered approach aims to protect information from sophisticated AI analysis.

Lu believes Ghost Font could have practical applications, particularly in CAPTCHA systems, which are increasingly vulnerable to AI bypass. By incorporating motion, the font could make it significantly harder for automated bots to solve challenges while remaining relatively easy for humans. However, Lu acknowledges that as AI technology advances, current methods may only offer a temporary solution, noting that the gap between AI perception and human readability continues to narrow.