How would you decide what is the middle ground though? If a project allows some AI-generated PR if its good quality, then it is a burden on the reviewer on what is considered good or not.
Maybe also limit the size/scope of external contributions (only small bug fixes allowed for your first few PRs)