I don't code as such any more, but the obvious two reasons I see is that coders would avoid it is for copyright issues as well as, since one doesn't know what the llm was trained on, for not unwittingly [1] introducing a crafty not yet known to the public exploit.

[1] Assuming that if LLMs were that good, all code would presently be double checked and exploits and back doors will shortly coming to an end - unless the reason it hasn't is a marketing issue.