Skip to content
AI Security

Interview question · Adversarial ML & Model Privacy

Accuracy drops 20 percent. How do you tell an attack from drift?

Drift is the world changing. An attack is someone changing it on purpose. The evidence that separates them is where and how the errors cluster.

mediumNewPreparation tags: CrowdStrike, Microsoft, GoogleUpdated Nov 2026

Drift is the world changing. An attack is someone changing it on purpose. The evidence that separates them is where and how the errors cluster.

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.

signing in unlocks more free answers per topic

What if · the follow-ups in this answer

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

A fraud model's detection rate drops sharply over two weeks, and the team must decide whether it is under attack or the data has drifted.

  1. What would you check first?
  2. How does natural drift usually look?
  3. How does an attack usually look?
  4. What if the errors cluster on a few new merchants?
  5. What do you do while investigating?
  6. How would you run the investigation?

Each rung has a nudge, a model answer, the principle that carries over and the weak answer to avoid. Sign in free to open it.

UP NEXT ON YOUR JOURNEY
Next in this trackExplain FGSM and PGD, and what the perturbation budget means.Next in this trackHow does adversarial training work, and what are its limits?Next in this trackWhat is gradient masking, and why is it a trap when you evaluate defences?
DISCUSSION · 0

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