they flag non-native English speakers for cheating more than native speakers.
Yes, and for me as a former it's absolutely clear why - because I'm doing the same thing as a generative model, imitating text in another language. Maybe with more practice in verbal communication and being more relaxed I could reduce this probability, but the thing is this is not something which should affect school tests at all.
These are people trying to use a specific kind of tools where it's fundamentally not applicable.