Skip to content
AI Security

Interview question · Adversarial ML & Model Privacy

How would you test whether a model update increased memorisation, and does a leak that needs thousands of queries matter?

Generic leakage benchmarks will not tell you whether your customers' records got easier to extract. Planted canaries, matched controls and an attack-economics view of severity will.

hardNewPreparation tags: Google DeepMind, OpenAI, AnthropicUpdated Nov 2026

Generic leakage benchmarks will not tell you whether your customers' records got easier to extract. Planted canaries, matched controls and an attack-economics view of severity will.

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 version of your assistant's model reproduces training text more often than the old one in internal testing, and the release is due next week.

  1. How do you measure memorisation consistently across versions?
  2. What else should the test include?
  3. What if the regression is in a high-sensitivity category?
  4. What if extraction needs thousands of adaptive queries?
  5. How would you fix the regression?
  6. What release decision process would you set?

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 trackHow would you red-team a content moderation classifier for evasion, and report results that are useful to the team that owns it?Next in this trackWhat are the main ways attackers 'break' a machine learning model, and which attack fits which level of attacker access?Popular right nowWhat security risks does the Model Context Protocol (MCP) introduce?
DISCUSSION · 0

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