This should have been rejected as an idea just on its face. False positives are really problematic. And if it performs unexpectedly well (accuracy is high) then

This should have been rejected as an idea just on its face. False positives are really problematic. And if it performs unexpectedly well (accuracy is high) then it just becomes a training tool for reinforcement learning.