Skip to content
AI Security

Interview question · Adversarial ML & Model Privacy

What is gradient masking, and why is it a trap when you evaluate defences?

A defence that breaks the attacker's tools is not a defence that stops the attacker. The scored answer explains how masking fools gradient attacks, how to spot it, and how adaptive attacks get around it.

hardNewPreparation tags: Google DeepMind, Meta, NVIDIAUpdated Nov 2026

A defence that breaks the attacker's tools is not a defence that stops the attacker. The scored answer explains how masking fools gradient attacks, how to spot it, and how adaptive attacks get around it.

Make your first attempt

What would you say before reading the answer?

  1. 01 State the assumptions your answer depends on.
  2. 02 Explain the mechanism and choose a control.
  3. 03 Name a condition that would change your decision.

The worked answer includes an approach, scoring criteria and follow-up questions. Compare the reasoning with your own.

10 answers per topic instead of 5, plus saved progress and bookmarks · costs nothingor unlock all 529 remaining answers · $25

Every worked answer in all 17 topics, every lesson in every course, the full concept curriculum, the complete practice-test bank, and the downloadable security field guides.

6 months · One payment · No auto-renewal

What if · the follow-ups in this answer

The answer works through 6 follow-ups like an interviewer would

A new defence shows near-perfect accuracy against PGD attacks, and the authors want to publish it as a major improvement.

  1. What is gradient masking?
  2. What signs suggest masking?
  3. How did Athalye, Carlini and Wagner break such defences?
  4. What if the defence uses random input transformations?
  5. What evaluation would you require before publication?
  6. How would you evaluate the claim?

Each rung has a nudge, a model answer, the principle that carries over and the weak answer to avoid. Premium opens every answer and ladder.

UP NEXT ON YOUR JOURNEY
Next in this trackA team claims their model is now robust to adversarial inputs. How do you evaluate that claim honestly?Next in this trackIf a model only returns accept or reject, how can an attacker still find inputs that evade it?Next in this trackHow would you red-team a fraud detection model?
DISCUSSION · 0

No comments yet. Share a correction or a different way to answer it.