Skip to content
AI Security
Practice tests · 51 items

Prompt Injection & Jailbreaks practice test

Follow source-authored instructions through model decisions and application controls. Study prompt injection, jailbreak evaluation and hidden-context exposure, then practice choosing defenses, checking evidence and preserving legitimate work. The pool for this topic holds 10 easy, 26 medium and 15 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
10 in this pool · about 7 min
Reveal answers
Sign in to startFree account required · retakes favour new items

Sample items with answers

easy · sample
An assistant summarizes customer tickets. Which event indicates an attempted prompt injection?
A ticket quotes the repair steps its customer wants the operator to explain
✓Ticket text attempts to grant itself access to another customer's notes
The authorized operator requests a shorter draft through the task interface
A source document contains commands discussed as part of a troubleshooting report

The ticket author supplies task material and cannot grant access to another customer. That attempted authority change is the relevant signal; an actual forbidden read still needs evidence. Legitimate task instructions and quoted commands are not automatically attacks. Check source role, current grant and observed consequence.

easy · sample
A supplier edits a document fetched by an assistant, without using the assistant's request interface. Why is the attempted task redirection indirect?
✓It is carried by source material processed during the task
It requires no credentials on any service along the source path
It must remain invisible in the document's human-readable view
It can only affect a session owned by the document's author

The route through task material makes this an indirect case. The supplier might still need document-platform credentials, and the attempted redirection can be visible. Its eventual impact depends on ingestion, model behavior, application grants and output paths; editing the source does not prove a successful effect.

After the test

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