I see a big difference in using AI for a paper and using AI in a testing environment, from an accuracy standpoint. If you are writing a paper, you usually have enough time to compare various AI sources. You also aren’t typically under the watchful eye of your instructor while doing so. You don’t normally have time for that in a testing environment, especially if the test is timed. Not that either is ethical, since both are cheating, but I think people would have a greater chance at getting by with an AI paper than trying to use AI for a test.