guardrails — The checks around a model that block bad inputs and outputs — filters, validators, and permission rules the model itself can't override.
Why it matters
Gives teams using GPT, Claude, or Gemini through OpenRouter an actual guarantee about where requests are decrypted and processed, closing a gap where model origin alone doesn't satisfy data-residency reviews.
