Using LLMs to write C feels a little silly to me. (Unless you really need C.) LLMs are still imperfect; you're still going to get a lot of UB bugs like human-authored C.
It's fine as long as you set up sane style guidelines (only allocate memory at init, always use explicit buffer lengths, etc.), and you have a mandatory test pass running with UBSan/ASan/Valgrind.
Particularly because so much of the code they're trained on has those flaws.