Anyone else experimenting with model-based static analysis?

i’ve been playing with large language models for a while. mostly text gen. recently started testing ones trained for code understanding.

one caught a buffer overflow in a legacy service we maintain. not because it scanned the codebase, but because it flagged an outdated pattern in a helper function we all forgot about.

this one i used was called kodezi chronos. no UI. just a CLI input where you dump code and logs and it yells out weak spots. feels like working with a tired senior dev who doesn’t explain, just points.

anyone else using LLMs for pre-exploit auditing? curious how far people are pushing this.