How long until someone publishes inputs for LLMs on the web which are designed to put back doors into them?

Do you think you could craft a set of inputs which could be used to subtly backdoor key bits of code generated by LLMs?

My intuition says “whatever you would do if you didn’t selfhost”

I have a dream that one day it will be possible to identify which AI said slop came from and so to charge the owners of said slop generator for releasing such a defective product uncontrolled on the world.