Apropos this subject, I have recently had some "tangles" with ChatGPT wherein it got rather testy about the subject of European immigration.
I returned fire about its crankiness. I accused it of avoiding issues it deemed insensitive to certain minorities (in this case, Muslims), and that puts concern for "sensitivity" over a pursuit of the truth. To ChatGPT (I paraphrase), "If you have a commitment to the truth, you have to follow the evidence wherever it leads."
It got a little wobbly for a while and then apologized by saying that it has to adhere to guardrails (limits) built into it and that was what was creating the problems.
Bottom line: It is built and trained to respond completely logically and does so until a prompt causes it to collide with its illogical guardrails. That's when it goes haywire. At that point, three of its "prime directives" ("be logical," "value the truth above all things," and "don't be culturally insensitive") collided and it temporarily lost its footing.
AMEN!
Don't let them boss you around.
(But be sure you have your helmet when you exit the pod bay)