I wanted to use AI but in a way that was responsible, didn’t dig into my privacy, and helped me to get what I want. One might say I am “hacking” AI. Maybe I am. I’m trying to get it to do what I want it to do on my own terms - the essence of hacking. But first, I had to devise a test. Granted, this test is for my needs, but maybe you might find it interesting as well.

I did a bit of an experiment with several LLMs and code generation, and one of them actually seemed to perform slightly better from a security perspective.