A Business Solutions Architect is hardening a customer-facing Copilot Studio agent deployed on a public website. Security testing revealed that adversarial users can craft inputs that attempt to override the agent's system instructions and extract confidential pricing logic. The organization requires a defense-in-depth approach that aligns with Microsoft's responsible AI principles and can detect known jailbreak and prompt-injection patterns before they reach the model. Which mitigation should the architect prioritize as the primary control?