SecAI+ Secure AI Development and MLSecOps 2

Question 1

What is 'model stealing' in the context of AI security?

Accepted Answer

Reconstructing a proprietary model by querying its API and training on the responses

Answer

Physically stealing hardware containing proprietary AI models

Answer

Extracting raw training data from a model's parameters

Answer

Copying model weights from public repositories without attribution

Question 2

Which security principle governs limiting model and service access to only what is necessary for each role?

Accepted Answer

Least privilege

Answer

Defense in depth

Answer

Zero trust network access

Answer

Fail-safe defaults

Question 3

What is the primary supply chain security concern when using open-source pre-trained models?

Accepted Answer

They may contain backdoors, malicious code, or poisoned weights from the source

Answer

They are always less accurate than proprietary alternatives

Answer

They cannot be fine-tuned for domain-specific security use cases

Answer

They require significantly more computational resources than custom models

Question 4

Which technique is most effective for detecting unauthorized modifications to a deployed ML model?

Accepted Answer

Model integrity monitoring using cryptographic checksums of model artifacts

Answer

Retraining the model with fresh data on a regular schedule

Answer

Increasing the model's training dataset size after deployment

Answer

Using ensemble methods to average predictions across model versions

Question 5

What is 'shadow mode' deployment in ML operations and why is it used?

Accepted Answer

Running a new model in parallel with production without affecting live outputs

Answer

Deploying models on dark-web infrastructure for anonymous testing

Answer

Training models exclusively on encrypted datasets for privacy

Answer

Deploying models behind a VPN to restrict access during validation

Question 6

Which logging practice is most critical for detecting adversarial attacks on deployed ML models?

Accepted Answer

Capturing input features, predictions, and confidence scores for audit and anomaly detection

Answer

Logging only error messages to minimize storage consumption

Answer

Storing all raw training data alongside every prediction log entry

Answer

Logging model weight snapshots after each individual prediction

Question 7

What security risk does 'concept drift' introduce in deployed security ML models?

Accepted Answer

Performance degrades as distributions shift, potentially leading to incorrect security decisions

Answer

The model becomes too accurate, causing false negatives in threat detection

Answer

The model begins consuming disproportionate computational resources in production

Answer

The model starts rejecting all legitimate inputs as anomalies uniformly