NCA-GENL Trustworthy AI Practice Question
Which THREE practices are recommended to minimize 'Data Leakage' in generative AI applications?
⚠ Common exam trap
Candidates often include 'model weight encryption' or 'increased training epochs,' which are security or training parameters that do not address the root cause of sensitive data entering the model.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Automated PII redaction during the data pre-processing phase.
Data leakage occurs when sensitive or private information is inadvertently included in training data or revealed through model outputs. Minimizing this requires PII redaction, strict access control, and output filtering. These measures are essential for Trustworthy AI because protecting user privacy is a foundational responsibility, and failure to control data flow can lead to significant legal, ethical, and reputational damage to an organization.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Automated PII redaction during the data pre-processing phase.
Why this is correct
PII redaction is the primary defense against data leakage. By scrubbing names, addresses, and other identifiers from the training corpus, developers ensure the model never learns to associate private data with patterns. This is a mandatory step for any system that interacts with sensitive user-generated content.
- ✗
Using public, unverified data sources to maximize training diversity.
Why it's wrong here
Using unverified data is a major source of security risk and potential leakage. Public data often contains unscrubbed private information, and using it without vetting can lead to the model incorporating sensitive data into its weights, which is exactly the opposite of the best practice for data protection.
- ✓
Implementing strict access controls for training datasets.
Why this is correct
Restricting who can access the training datasets prevents unauthorized exposure of data before it even reaches the model. This is a critical security layer that ensures data is only handled by authorized personnel, which is a standard procedure for maintaining the integrity and privacy of the training pipeline.
- ✓
Deploying output filters to detect and block PII in real-time.
Why this is correct
Even with sanitized training data, there is a risk of a model hallucinating private information or exposing data it should not have. Real-time output filtering serves as a final safety net, checking for sensitive patterns before they reach the user, adding a vital layer of defensive security.
- ✗
Increasing the learning rate to ensure faster convergence and data masking.
Why it's wrong here
Learning rate is a hyperparameter for optimization and has no relationship with data security or privacy. It cannot 'mask' data. Confusing optimization parameters with security controls is a fundamental misunderstanding of model training; learning rate adjustments have no impact on the privacy or security of the data.
About these practice questions
This NCA-GENL question is part of Courseiva's 367-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official NVIDIA exam blueprint
This NCA-GENL practice question is part of Courseiva's free NVIDIA certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the NCA-GENL exam.