Adaptive training improves AI text detection in tests, researchers report
MIT and Cornell researchers tested a method that lets deployed detectors adjust to new, unlabeled writing. It improved results against disguised AI text and simulated changes in human language, though the researchers caution that future detection may become harder.











