We're already at that point. It's already impossible to tell if any given function or class was written by AI. You can only tell when the code is large; where the architecture reached the point where AI couldn't keep it all inside the context window.
Eventually the AI will write better code than humans and that's when policies like this will seem very short sighted.