Question 100
Original unofficial Palo Alto Networks NetSec-Architect scenario: A retail chatbot is already live and receives user prompts containing attempts to elicit restricted customer information. Which continuing assurance practice best prevents recurrence?
Explanation
Correct Answer
Original fictional security architecture situation: A retail chatbot is already live and receives user prompts containing attempts to elicit restricted customer information. Governing principle: Live prompt and response risk exists after model approval and requires runtime enforcement. The most defensible design response is: Place a supported AI runtime security control in the inference request and response path, backed by tested sensitive-data rules. Validation: A synthetic prompt trying to reveal protected data is blocked or safely transformed and recorded. Sustained assurance: Monitor the impact of runtime guardrail rule changes on legitimate conversations
Incorrect Answers
“Include tool-result injection tests in release acceptance criteria.” concerns the separate design topic: differentiate prisma airs and ai access capabilities. It would not fix the described weakness: a retail chatbot is already live and receives user prompts containing attempts to elicit restricted customer information. Live prompt and response risk exists after model approval and requires runtime enforcement.
The suggested “Inventory direct API clients before claiming full AI usage coverage.” tests a different architectural concern: select ai security reference architectures, form factors, and data controls. Here, a retail chatbot is already live and receives user prompts containing attempts to elicit restricted customer information. Live prompt and response risk exists after model approval and requires runtime enforcement.
The proposed design “Test bidirectional inspection and retrieval permissions as one control set.” addresses this related architecture topic: select ai security reference architectures, form factors, and data controls. The actual security gap is that a retail chatbot is already live and receives user prompts containing attempts to elicit restricted customer information. Live prompt and response risk exists after model approval and requires runtime enforcement.