AISG practice · Module 6

Secure AI Architecture, Guardrails & Agentic IAM — 25 questions. Choose the BEST answer.

You can answer as a guest, but you'll need a free account to see your score and save it. Sign up · Sign in
Question 1 Question 1 · M6
Which option gives an organisation the MOST control but also the MOST operational responsibility?
Question 2 Question 2 · M6
What does ISPM help an organisation find?
Question 3 Question 3 · M6
Which identity should an autonomous agent use?
Question 4 Question 4 · M6
An MCP server forwards the user's access token unchanged to a third-party API. What is the MAIN risk?
Question 5 Question 5 · M6
What is over-delegation?
Question 6 Question 6 · M6
Streaming model output token by token to users primarily complicates which control?
Question 7 Question 7 · M6
An external scan finds an unauthenticated model-server management API. What should be done FIRST?
Question 8 Question 8 · M6
Which mechanism lets workloads authenticate without stored long-lived secrets?
Question 9 Question 9 · M6
An agent that proposes vendor payments can also approve them. Which principle is violated?
Question 10 Question 10 · M6
Why should autoscaling for inference have a ceiling?
Question 11 Question 11 · M6
What does mutual TLS add to service-to-service communication between agents?
Question 12 Question 12 · M6
Which approach BEST isolates multiple tenants sharing GPUs?
Question 13 Question 13 · M6
Which authorisation model BEST suits per-tool-call decisions in an agent harness?
Question 14 Question 14 · M6
A model-provider API key is found in a notebook committed to a repository. What should happen FIRST?
Question 15 Question 15 · M6
A local MCP server is requested for a coding assistant. What is the MOST important containment?
Question 16 Question 16 · M6
What is context boundary validation?
Question 17 Question 17 · M6
Notebook servers in a research project hold long-lived cloud keys. What is the BEST fix?
Question 18 Question 18 · M6
Applying Zero Trust to an AI application MOST directly means:
Question 19 Question 19 · M6
Which test type checks behaviour beyond expected limits, such as exhausting context windows and GPUs?
Question 20 Question 20 · M6
What should happen when model output fails schema validation in an agent workflow?
Question 21 Question 21 · M6
Logs show actions from 'priya@bank' but cannot tell whether Priya or her assistant agent acted. What is missing?
Question 22 Question 22 · M6
A design relies solely on the system prompt instructing the model to ignore injected text. Which principle does this MOST clearly violate?
Question 23 Question 23 · M6
What is tool poisoning in MCP?
Question 24 Question 24 · M6
In MCP, what is a 'rug pull'?
Question 25 Question 25 · M6
Why should functional tests of an LLM feature run each scenario several times?