Artificial Intelligence Attempts to Protect Texts from Itself!

In an attempt to reverse the roles between humans and machines, the type design company Mix Font has created an experimental font that allows human readers to see one message while AI systems may read a different, misleading one.

According to the Smithsonian website, the innovation is called “Decoy Font” and relies on a visual trick that combines two characters within the same space. One character appears with thin, clear lines, while the other is formed from a dark, low-detail mass. The human brain can distinguish between the two layers, especially when viewing distance changes, but some image-reading models focus on the clearer layer and fail to extract the hidden message.

The font is presented as an experiment to protect human writing from automated collection programs and text extraction tools. It can be downloaded in TTF format and used within design software.

However, this method does not provide guaranteed protection and is not a substitute for encryption. A system that accesses the original digital text can read it directly, and newer AI models may bypass the trick if they are previously aware of how it works.

Independent tests indicate that providing a simple hint to AI may help it uncover the true message.

Therefore, the concept appears closer to an experiment in the differences between human and machine perception, and to a temporary race between deception design and the development of computer vision, rather than serving as a permanent digital lock.