SecAI+ Secure AI Development and MLSecOps 1 — Questions and Answers
Question 1: What is MLSecOps primarily focused on?
- Integrating security practices into the ML development lifecycle (Correct answer)
- Using ML exclusively to detect network intrusions
- Replacing traditional DevOps pipelines with AI automation
- Automating vulnerability scanning using neural networks only
Correct answer: Integrating security practices into the ML development lifecycle
MLSecOps integrates security practices throughout the machine learning development lifecycle, from data collection through model deployment and monitoring.
Question 2: Which phase of the ML pipeline is most vulnerable to data poisoning attacks?
- Model inference and serving
- Model deployment to production
- Training data collection and preparation (Correct answer)
- Model evaluation and benchmarking
Correct answer: Training data collection and preparation
Training data collection and preparation is most vulnerable because poisoned data injected at this phase corrupts the model's learning and produces malicious behavior at scale.
Question 3: What is the primary purpose of model versioning in a secure ML pipeline?
- To accelerate model training iterations
- To track changes and enable rollback if a model is compromised (Correct answer)
- To share models publicly for community transparency
- To reduce inference computational costs over time
Correct answer: To track changes and enable rollback if a model is compromised
Model versioning enables teams to track all changes over time and roll back to a previous, uncompromised version if a security incident is discovered in the current deployment.
Question 4: Which practice best ensures the integrity of ML artifacts during a CI/CD build pipeline?
- Randomizing training data splits before each build
- Cryptographic signing and hash verification of model files (Correct answer)
- Using larger training datasets for every release
- Disabling model logging during training to reduce overhead
Correct answer: Cryptographic signing and hash verification of model files
Cryptographic signing and hash verification ensures ML artifacts have not been tampered with during storage, transfer, or the build process — the same principle used in secure software supply chains.
Question 5: What does a 'model card' document primarily provide in secure AI development?
- A legal software license governing AI system usage rights
- Transparency about a model's capabilities, limitations, and intended use cases (Correct answer)
- A performance benchmark comparing the model against competitors
- An API specification for integrating the model into applications
Correct answer: Transparency about a model's capabilities, limitations, and intended use cases
Model cards provide transparency about training data, intended uses, known limitations, and potential biases, supporting responsible deployment and security review processes.
Question 6: In a secure ML pipeline, what is the primary security role of a feature store?
- A database storing only raw, unprocessed data for compliance
- A centralized repository for managing, versioning, and sharing ML features consistently (Correct answer)
- A tool for visualizing model predictions in real time
- A component that encrypts model weights during inference requests
Correct answer: A centralized repository for managing, versioning, and sharing ML features consistently
A feature store provides centralized, versioned, and consistently governed management of ML features, preventing inconsistencies and unauthorized feature modifications across teams.
Question 7: Which security control is most critical when exposing ML models as REST API endpoints?
- Using the largest available model size for accuracy
- Authentication, authorization, and rate limiting on model endpoints (Correct answer)
- Making model weights publicly available for third-party audits
- Disabling request logging to protect user privacy from the start
Correct answer: Authentication, authorization, and rate limiting on model endpoints
Authentication, authorization, and rate limiting protect model API endpoints from unauthorized access, abuse, model extraction attacks, and denial-of-service conditions.
SecAI+ Secure AI Development and MLSecOps 1
What is MLSecOps primarily focused on?