SecAI+ Secure AI Development and MLSecOps 3

Question 1

What is the primary security benefit of a 'canary deployment' strategy for ML model rollouts?

Accepted Answer

It gradually routes a small traffic slice to the new model to detect issues before full deployment

Answer

It provides bird-themed model naming for easier categorization

Answer

It deploys models exclusively in isolated cloud environments for security

Answer

It tests models only on synthetic data before any production exposure

Question 2

Which framework specifically addresses risk management across the machine learning and AI lifecycle?

Accepted Answer

NIST AI Risk Management Framework (AI RMF)

Answer

OWASP Top 10 for Web Applications

Answer

PCI DSS for payment system security

Answer

ISO/IEC 27001 for information security management

Question 3

What is 'data lineage' and why is it essential in a secure AI development environment?

Accepted Answer

A record tracking data's origin, transformations, and usage throughout the ML pipeline

Answer

The geographic origin of training data, used for regional regulatory compliance only

Answer

The chronological order in which model versions were trained on successive datasets

Answer

A method to anonymize sensitive data before it enters model training

Question 4

What is the primary security purpose of applying differential privacy during ML model training?

Accepted Answer

To prevent individual data points from being inferred from the trained model

Answer

To accelerate model convergence and reduce training time

Answer

To generate multiple model variants for A/B security testing

Answer

To balance computational load evenly across distributed training nodes

Question 5

Which access control model is most appropriate for a multi-tenant ML platform serving multiple business units with different data sensitivity levels?

Accepted Answer

Role-Based Access Control (RBAC) with attribute-based policy extensions

Answer

Discretionary Access Control (DAC) managed by dataset owners

Answer

Mandatory Access Control (MAC) with fixed security labels

Answer

An open access policy with manual approval workflows for all requests

Question 6

What is a 'model inversion' attack and what does it attempt to recover?

Accepted Answer

Exploiting a model's outputs to reconstruct sensitive training data examples

Answer

Reversing a model's predictions to identify mislabeled training examples for quality

Answer

Inverting the weights of a neural network to produce a compressed model variant

Answer

Running a model in reverse to automatically generate adversarial input samples

Question 7

Which practice best mitigates ML supply chain attacks when integrating third-party ML libraries?

Accepted Answer

Pinning specific library versions and verifying their integrity with cryptographic checksums

Answer

Always downloading the latest library version at each pipeline execution

Answer

Using only closed-source commercial libraries to avoid open-source risks

Answer

Running all ML pipeline code with administrator privileges for compatibility