GitHub Certified: Agentic AI Developer

0/13 checked

Q_01 domain/d1 · medium single discuss →

Your team is designing a RAG-enabled agent that retrieves documents from Azure AI Search and passes them to the model. During architecture reviews, a security engineer asks how retrieved content should be handled to keep the SDLC secure. Which design principle should you adopt?

A Treat retrieved content as trusted since it comes from your own index, and skip injection mitigation
B Treat retrieved content as untrusted input and apply measures against indirect prompt injection
C Strip all source metadata before passing content to the model to reduce token usage
D Disable index filters so the agent can access the broadest possible knowledge

Q_02 domain/d1 · medium single discuss →

You are preparing the SDLC evaluation stage for an enterprise agent platform. The team wants to know how agent risks such as cross-prompt injection attacks and ungrounded outputs were assessed before release, based on the Agent 365 responsible AI guidance. Which evaluation approach was described?

A Only automated unit tests on the underlying model weights
B A variety of assessment activities including AI red teaming and evals that identified defect rates
C A single manual code review by the development team
D Solely customer-reported feedback collected after general availability