This is probably a tiny amount of the guardrails. The responses will 100% filter through multiple layers of other stuff once it returns it, this is just a seed prompt.
They also filter stuff via the data/models it was trained on too no doubt.
They also filter stuff via the data/models it was trained on too no doubt.
It’s a lot less complicated than you would be lead to believe