This thought process begs the question, why not just review code and deny what's bad, rather than blanket ban any code with LLM involvement? That way you retain the productivity benefits of LLMs, and can still deny flawed code.