What You See Is Not What They Get

What You See Is Not What They Get focuses on practical research into defense mechanisms against AI data scraping and agents, relying not on access prevention methods or obfuscation of provided data, but rather on exploiting the discrepancy between machine reading and human visual perception.

For the residency, Fontu, a Python tool, is being developed. It builds the foundation, taking HTML files as input, and encrypts the content of text elements within it by scrambling the letters into it. The result is valid HTML, but its text content is illegible to machines. Decryption is not performed by software, but by the eyes of the website’s visitors.

During encryption, Fontu creates a specific version of a font in which the characters are swapped in exactly the opposite order than they appeared in the HTML file’s text. If letter A was replaced by letter C, then the generated font displays char C in the shape of an A.

What You See Is Not What They Get also addresses accessibility by exploring possibilities of providing audio transcripts for visually impaired people with properly placed ARIA (Accessible Rich Internet Applications) elements that would complement the font.

The project aims to address OCR (Optical Character Recognition) attacks, investigating the creation of font shapes to resist OCR through test-driven research, and rapid blending of multiple font shapes into a readable letter outline. Another aim of automatonomy.net’s project is improving encryption, by exploiting ligatures and substitution to unused characters.

The project will result in an open-source code, a project website featuring examples, user guides, links to open-source repositories containing the source code, and comprehensive a documentation of the development process covering the concept, code, and visual design of the project. The entire project will be made available to the community for further development and use.