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?

An e-commerce platform's AI assistant has been manipulated by customers who discovered that certain user prompt patterns can extract the system's internal pricing algorithms and discount rules. Investigation shows the system prompt contained these rules for context, making them vulnerable to extraction. What fundamental system prompt security principle was violated?

A data protection officer configures the company's AI platform to inspect all prompts for compliance violations and block queries exceeding 4,000 tokens per request. Which two gateway controls are simultaneously enforced?

An organization deploys an AI-driven response system that prioritizes and automatically assigns alerts based on severity, containment status, and affected assets. Which use case does this describe?

An AI fraud detection model becomes unresponsive after attackers send excessively large and malformed data payloads, causing GPU memory exhaustion. Which two attack types best describe this event?