A financial institution uses hashing and cryptographic validation to verify that training datasets have not been tampered with before model retraining. Which compensating control is applied?

A financial chatbot undergoes pre-deployment evaluation to ensure it never produces investment advice or discloses personal data. The tests include adversarial prompting and stress validation. Which two controls best describe this testing?

A data validation team reviews AI-generated legal summaries and cross-verifies them against official documents to ensure the information aligns with established facts. What quality metric is being audited?

An enterprise uses AI in its CI/CD workflow to simulate various input data scenarios and evaluate how updated ML models respond compared to the baseline model. Which two testing types are combined in this process?

A cloud-based AI service enforces user authentication tokens before allowing model access. Each endpoint has separate permissions defining which users can submit prompts or retrieve results. What mechanism does this describe?