Guide-linked quiz · Hard Level
Prompt Injection Detection Classifiers Quiz
Tests classifier scope, threshold tradeoffs, Prompt Guard 2 version limits, Microsoft Prompt Shields configuration, and defense-in-depth.
Question 1 of 8
What does an injection detection classifier return?
Preview every question8 questions
Answers and explanations appear as you take the quiz.
- What does an injection detection classifier return?
- A label or score about whether text resembles an attack under its policy
- A guarantee that the input is safe
- A corrected system prompt
- Proof that the main model would obey the text
- What kind of text does Meta’s Prompt Guard 2 model card target?
- Only generated model outputs
- Explicit attempts to override prior instructions, labeled benign or malicious
- All false statements and factual errors
- Every harmful image
- What context-window limit does the Prompt Guard 2 card list?
- One million tokens for every detector
- A limit only on image inputs
- 512 tokens for the documented models
- No input limit
- Which risk comes with setting an injection-classifier threshold too aggressively?
- Novel attacks will always be blocked
- The system stops needing evaluation
- All benign content is classified correctly
- Legitimate security discussions may be flagged
- What can happen with a threshold that is too permissive?
- Novel or reworded attacks may be missed
- Every benign input is blocked
- The classifier turns into a language model
- The tool permissions change automatically
- What does Microsoft document about Prompt Shields configuration?
- It guarantees no false positives
- Intervention points must match where the attacks appear
- The feature scans every deployment automatically
- It cannot flag document attacks
- Why should a team record classifier version and intervention point?
- To avoid tracking false positives
- To bypass all model limits
- To interpret and reproduce why particular content was scored or filtered
- To make the score a legal conclusion
- How should classifier results fit into a security design?
- As the sole proof that all unflagged text is safe
- As a replacement for testing
- As an automatic authorization for tools
- As one signal combined with provenance, permissions, validation, and monitoring
Keep testing yourself
More quizzes picked for your level.