Curious: how do you exactly detect an AI-generated comment?

A case when security through obscurity is perfectly justified.