I've noticed there seems to be a default style that is easier to detect. I've noticed it harder to detect when asking an LLM to use a different style (more conversational, avoid sounding like an AI, don't use emdashes, etc). I wonder if that's what you're picking up too - the instances where people make no effort to change the style of the output.

How do you estimate your false negative rate?

What's likely to happen though, is human idiomatic writing will degrade to AI level and the two will converge. Just like nowadays it is harder to tell if some (human) non-native English speaker wrote some English text online whereas 20 years ago it was very obvious.