Despite growing opposition from the public and some lawmakers, there’s still not much stopping tech companies from training their generative AI programs on Internet content without the author’s consent. Therefore, until meaningful legislation and oversight is introduced, people will largely be on their own when it comes to protecting original material online.

While designer Eric Lu’s Ghost Font isn’t necessarily the silver bullet needed to stop AI from completely devouring the internet, it’s still an ingenious proof-of-concept experiment that highlights human creativity. It also shows how far AI has to go when it comes to analyzing concepts like optical illusions.

Ghost Font’s underlying mechanisms are deceptively simple. A user enters text and overlays it within hundreds of moving dots in a short video animation. The dots that make up the words go up on the screen while the rest of the dots move down, generating a visual opposition that the human mind can detect and discern. Pausing the video creates a stationary image that allows the words to disappear amid the general clutter. AI (or the human eye, for that matter) can’t decipher the message in the still image, but that’s also where people get the upper hand.

Ghost Source Anti-AI Demo

While the human brain interprets motion as a fluid pattern, multimodal AI still examines videos based on their individual frames. Think of it like a flip book. Humans will see the cumulative animation as they scroll through the pages, but the AI ​​is still forced to look at the image on each individual page. As added protection, Ghost Font files also include the static decoy message “written in Ghost Font”, which further confuses the AI ​​by making it appear that that is the only real text to scan.

Lu claims to have tested the program using Anthropic’s Claude Fable and OpenAI’s GPT-5.6 Sol Ultra, both of which failed to find the actual text. Other tests conducted by Creative Bloq also seemed to be better at AI.

By Lu’s own admission, Ghost Font is not a panacea for current AI problems. It is limited to extremely short messages and may become obsolete if generative AI develops the ability to examine videos as optical flows rather than individual frames. Plus, it only takes a few seconds watching a dizzying Ghost Font video to realize that it’s not easy on the eyes and is probably useless for anyone who is visually impaired. Regardless, it’s still a fun, free way to celebrate human ingenuity and AI shortcomings.

Disclosure: Ziff Davis, the parent company of Popular Science, filed a lawsuit against OpenAI in April 2025, alleging that it infringed Ziff Davis’ copyrights by training and operating its AI systems.

2025 PopSci Best of what’s new

The 50 most important innovations of the year