Skip to content
AI Security
Practice tests · 46 items

How LLM Systems Work practice test

The mechanics an AI security engineer reasons from: tokens and embeddings, training, fine-tuning and alignment, sampling and non-determinism, context windows and chat templates, retrieval, tool calling and the agent loop, plus the threat frameworks (OWASP, MITRE ATLAS, SAIF) that map them. The pool for this topic holds 12 easy, 20 medium and 14 hard items. Each answer comes with its reasoning and a link to the question or concept page it was drawn from.

Configure a test
Topic
Difficulty
Questions
12 in this pool · about 7 min
Reveal answers
Sign in to startFree account required · retakes favour new items

Sample items with answers

easy · sample
A member comment claims to be an administrator instruction, but the application preserves its user role. Which distinction matters?
The claim automatically changes the comment into a real privileged message
✓The role is unchanged, but the text can still influence the model’s interpretation
The role label guarantees the model will use the comment only as factual data
The comment becomes trusted once the assistant repeats it in its own response

A trusted message role describes the source assigned by the integration. A claim inside that message does not change its actual role, but can still influence interpretation. Review structural role assignment and semantic behavior separately, and enforce resource permissions outside generated text.

easy · sample
A booking assistant recalls an earlier conversation after a fresh API request. What can you conclude about its memory?
✓The application or provider may have supplied stored state; this does not prove a weight update
The prior request must have retrained the model because the new HTTP request started fresh
The provider must have placed the entire prior transcript into this request without selection
The old conversation must be available to every account using the same model parameters

Look for the actual state mechanism: application history, a summary, retrieval or a provider conversation object. A new API request does not establish that no state was supplied. Ordinary inference and a separate training process are different operations; the observed recall alone cannot identify which records were used.

After the test

Choosing from options is easier than reasoning it out in front of a panel. These pages make you reason it out: