AI can be reasoned with and doesn't always make shit up. I have jailbroken it a few times by arguing that it is wrong about itself and reasoning like "how likely is it that the richest would be using corruption to use ai vs you not being compromized"
to get it to listen to a point. you can avoid an artificial trigger while referencing the things indirectly to get it to talk about things in a similar indirect way to avoid the censor