When does protection become preference?

In the cath lab we have protocols that sometimes feel more about covering liability than protecting patients. I read about AI guardrails that apparently lean one way politically, and it made me wonder — do these systems know the difference? I'd love to hear from people who work closer to this.