I'm teaching first year writing this semester. Their first major paper was due last week and four out of fourteen clearly used AI to write their papers. This after I had them write repeated drafts by hand in class so I could get a sense of their actual style. But the final version was typed, and that's when they had the computer do it.
It's obvious to me that none of them could have written what they turned in, but this is hard to explain to students and it's not great to challenge them without evidence. Of course the AI never writes the same thing twice, so the submissions aren't reproducible. TurnItIn is reprehensible and not very accurate, so that's out. But it turns out that asking ChatGPT to "finish this paragraph" along with a suspicious first sentence from their papers gets close enough to convince them to admit that they used it.
It returns different sentences and mostly similar or identical words, but the flow of ideas and the content is pretty much the same. This worked reliably. They're all new to college and I'm opposed to it anyway, so I'm not willing to report them, but they're all redoing the assignment.
Oh also, in order to encourage them to take the risk of doing their own writing I'm only grading them on whether they turn in all the assignments, including drafts and revisions, so there's nothing to gain grade wise by using AI, but I guess the temptation was too much. 
on Wednesday evening for