LLM Application Security Specialist Diagnostic

A focused diagnostic for practitioners building or securing LLM-powered applications, RAG systems, and agentic workflows.

60 q · 75m

Question 1 of 602%

secure architecture design

Why is a content-safety filter not enough as an AI security architecture?

A customer-facing assistant has a moderation filter for harmful language. It also retrieves confidential documents, calls CRM tools, stores prompt logs, and routes data to model providers.

A. Blocking harmful language alone does not secure retrieval boundaries, logs, provider routing, or tool execution.B. Moderation filters reduce harmful language, but they do not enforce authorization, retrieval scope, or high-risk action controls.C. Content safety covers one risk class; security architecture must also enforce data, tool, routing, and identity.D. Model refusals help, but they cannot replace deterministic application-side control enforcement.