I love the idea of two AIs working against each other. One with the instruction to attempt to deny the claim, the other instructed to get the claim approved. In the end it is just a battle of who will pay more money to keep trying. And then, we can setup rogue bots they file random claims with some cheap opensource model just to burn the insurance companies tokens.