> They already doing that with AI, rejecting claims at higher numbers than before

Source?

Haven't risk based models been a thing for the last 15-20 years ?