During a governance review, a compliance officer asks how Microsoft's responsible AI approach ensures that Microsoft 365 Copilot behaves consistently and avoids producing harmful or unexpected outputs even under unusual prompts. Which responsible AI principle most directly addresses this concern?