After retiring at 65 from a university teaching and research science career (all pre-AI), I went back to teaching, but this time teaching high school science, mostly AP STEM courses at an A-ranked public high school. The cheating/AI problem is now a crisis greater than COVID. My experience: very few students in advanced and AP classes do not cheat — largely for the reasons given above — and it takes enormous resourcefulness on the teacher's part to design coursework and examinations in which cheating through AI is not an issue. Many teachers I know have all but given up — the cost and effort required to circumvent cheating are simply too great given the already sky-high demands on teachers' time and energy. And school administrations are little help, due to thoughtless and enthusiastic reliance on software at every level. In some ways they are part of the problem. I don't know what the situation is in schools outside the US. But here it had become an arms race.
[Edit: typos]
My daughter just finished her Grade 11 finals in Canada. They were done on locked down school Chromebooks, which should be enough to prevent cheating by all but the most dedicated.
Personally I believe AI has made exams and high stakes testing unworkable. Even before AI I would argue teaching to the test made high stakes testing unworkable. How grades are assigned IMO will be more like how employees are evaluated in the workplace: some metrics, some oral exams, some peer feedback, but mostly on what they produced.