Knowledge check

Quiz

In Progress

Which of the following best illustrates a threat to service integrity in an ML system?

A. A model registry is stored in encrypted cloud storage.B. A model deployed with public API access is modified through unauthorised queries.C. A team member accesses a model file without approval.D. A model's training data includes personally identifiable information.

Which infrastructure setup provides the strongest support for strict data residency requirements?

A. Fully cloud-based deployment in a multi-region setup.B. Federated learning using edge devices.C. Hybrid deployment with cloud-based training.D. On-premises infrastructure.

What is a major security challenge unique to federated learning?

A. Gradient updates may leak private information.B. Model rollback is difficult to implement.C. It does not support large-scale training.D. All training occurs on a single device.

Which of the following reflects a defence-in-depth strategy during the deployment phase?

A. Using a single firewall for the model endpoint.B. Avoiding model version control in production.C. Requiring API authentication and validating inference inputs.D. Running the model on a local device with no logging.

Why is zero trust an important principle in ML security?

A. It eliminates the need for encryption.B. It assumes internal systems are always trustworthy.C. It reduces reliance on cloud platforms.D. It ensures every request is authenticated and verified.

Which step should come first when applying threat modelling to an ML workflow?

A. Rate the impact and likelihood of each threat.B. Define access control and encryption policies.C. Map the ML workflow and identify assets.D. Respond to detected anomalies.

A platform engineer should be primarily responsible for which of the following tasks?

A. Selecting training hyperparameters.B. Reviewing patient consent forms for data use.C. Configuring access controls and monitoring endpoints.D. Performing exploratory data analysis.

Which technique is best suited for continuous security monitoring in ML workflows?

A. Using anomaly detection to track model behaviour.B. Encrypting model outputs.C. Reviewing access logs monthly.D. Allowing unrestricted endpoint access for easier debugging.

Why is it important to hold security post-mortems after ML incidents?

A. To document lessons learned and improve protocols.B. To assign blame and enforce strict penalties.C. To avoid involving the broader team in future incidents.D. To stop using CI/CD pipelines for deployment.

A junior team member accidentally deletes a production model. What security measure would most directly have prevented this?

A. Enabling model version control.B. Using federated learning instead.C. Encrypting the training data.D. Assigning least privilege access rights based on role.

Evaluate your understanding of this unit by completing the Knowledge Check.##

Action item: Knowledge check

In progress