LLM Poison
After being inspired by this blog post, I've set up my own humble little LLM poison trap on this site. It's not as sophisticated as the one you'll find in the page linked prior because of the limitations of Codeberg's static page hosting, but it was still a fun project to do. A link to the pages can be found here.
I implemented it by writing a Perl script that reads text from input, builds
a Markov tree of said input, then outputs a Markov-chain-generated output. I
used an aggregation of Slackware's fortune databases, Moby-Dick,
Three Men In A Boat, and Pudd'nhead Wilson as the source input text. Another
Perl script then takes these generated texts and converts them to HTML pages
which all link to each other, so that LLM scrapers will get "stuck" hopping
between these pages.
It was a fun project! I don't know how effective it actually will be considering its just 30 different static web pages, and Codeberg works hard to block LLM scrapers from scraping its services, but it feels nice to do my part in combatting the scourge of evil, incompetant scrapers taking over the web :-).
Links
- https://maurycyz.com/misc/the_cost_of_trash/ - Article that inspired me to pursue this endeavor.
- https://en.wikipedia.org/wiki/Markov_chain - Wikipedia article for Markov chains.
- https://www.dcode.fr/markov-chain-text - Online Markov chain text generator you can use to get an idea of what generated text will look like.
Last Modified: 2026-10-18