Why combine model-level safety with application guardrails?

Question

Accepted Answer

Defense in depth: models can fail; app policies, filters, and auth enforce hard constraints — Model safety + app guardrails = belt and suspenders. Either layer can fail alone.

Answer

Application guardrails are illegal

Answer

Model safety is always perfect alone

Answer

Guardrails increase prompt injection success

Why combine model-level safety with application guardrails?

Short Explanation

Full Explanation