"Guardrails" for AI have no more impact on its responses than a prompt. To make effective guardrails, you would need an AI that can actually think, or manually program in permutations of what you don't want it to say and do above the prompt level.
add a skeleton here at some point
6 days ago