Skip to content
AI Security

Interview question · Cloud, Container & AI Infrastructure Security

Design controls that protect model weights from external attackers and insiders.

One copy is a total loss. The design is about how few places weights exist, how few people can touch them, and whether a bulk copy can leave unnoticed.

expertEssentialNewPreparation tags: Anthropic, OpenAI, Google DeepMindUpdated Nov 2026

One copy is a total loss. The design is about how few places weights exist, how few people can touch them, and whether a bulk copy can leave unnoticed.

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 lab's most valuable asset is a set of model weights worth years of compute, and copies exist in several environments.

  1. Where do you start?
  2. How do you reduce exposure?
  3. How should encryption help?
  4. What about exfiltration?
  5. How do you know the controls work?
  6. How would you protect the weights?

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 trackHow would you control egress and detect exfiltration from a training cluster?Next in this trackWhy are exposed notebooks and distributed compute dashboards dangerous, and how do you prevent it?Next in this trackHarden a Kubernetes cluster that runs GPU training jobs.
DISCUSSION · 0

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