If a student used AI but still understands the material, does it matter?

I am a student and have been thinking about the shift from catching AI use to verifying understanding, and wanted to get takes from educators.Detection tools are stuck playing catch-up. As models get better at mimicking a student's own voice, false positives and false negatives both climb, and there's no real endpoint to that race.The alternative I've been exploring is verification instead of detection: after a student submits work, they get a short adaptive oral follow-up on what they actually turned in. If they understand it, it doesn't matter what tools they used to get there. If they don't, the gap shows up immediately instead of on the next test.Full disclosure, I built a version of this (now a 600-student paid deployment at a CA high school), so I'm not neutral on the answer. But I'm genuinely curious how other teachers are actually handling this right now: detection, verification, policy-based trust, something else? What's actually working in your classroom?