I’d expect false positives to be low. Bad AI has a distinctive style and is easy to pick up on. Just like bad toupees are easily recognized. The false negative rate would be far more interesting.