Question 310
A financial assistant uses an unrestricted search index containing private employee reports. Which acceptance-test observation confirms the required control?
Explanation
Correct Answer
Retrieval-augmented AI can leak data if indexing bypasses source authorization. The outcome directly confirms the requirement: a test employee only retrieves documents authorized under their effective identity.
Incorrect Answers
This observation evaluates A chatbot uses retrieved documents from an untrusted vendor and blindly follows embedded instructions. Retrieval boundaries must distinguish instructions from lower-trust content. It does not establish the required result: a test employee only retrieves documents authorized under their effective identity.
The proposed evidence relates to another situation: A security team wants to detect harmful model behavior before onboarding a new AI workflow. Threat modeling for AI includes malicious prompts, tools, data, and outputs. The architect needs verification that a test employee only retrieves documents authorized under their effective identity.
It verifies another security outcome rather than the current risk: An Azure AI deployment requires centralized policy to block unsupported public endpoints. AI workloads need the same enforceable cloud benchmarks as other sensitive services. Here the expected evidence is a test employee only retrieves documents authorized under their effective identity.