CYBRARY Unit 3 assessment — Adversarial ML and red-team operations
Attacks on the model itself — extraction, membership inference, evasion — and how a campaign is run and reported against an AI system without turning it into anecdotes.
6 questions drawn from a bank of 9, in their own order with the options moved. Take it again and you sit a different paper.
0 of 6 answered
1. Which single control best supports an AI security program across all these attacks?
Tests: Detection coverage for ML attacks
2. What makes an AI red-team result reproducible?
Tests: Reproducibility in AI testing
3. Where does poisoning sit relative to the others?
Tests: Poisoning versus runtime attacks
4. Membership inference asks which question?
Tests: Membership inference
5. An evasion attack changes a few pixels and flips the classification. What does that show about the decision boundary?
Tests: Evasion and decision boundaries
6. A red-team campaign report says 'the model can be jailbroken'. Why is that finding weak?
Tests: Reporting AI red-team findings