If you have 50,000 employees and are screening for a risk that is 1 in 1M with a 5% false positive rate you are going to be very disappointed when over the next decade it identifies 25,000 would be shooters when you have zero actual active shooters. Even better you will probably stop disregarding such a test and miss if if it actually happens.
As awesome the fact that skynet is always watching will probably cause people to manage their workspace personas to a psychotic degree that will surely ratchet up workspace stress to new highs. Deprived of actual data on what triggers the eye of sauron 100 wrong theories about how to avoid doing so will proliferate and your studied population will both diverge from the norm the system was designed to operate on and become progressively worse.
A few years later a study will prove that the AI inadvertently learned to discriminate against minorities, women, or people in other time zones through things the training population did without thinking and the people pushing it will look like bigots. Instead of ejecting we will try to fix it. Either this doesn't work or if it does people accuse skynet of being woke.