NCP-GENL Safety, Ethics, and Compliance Practice Question
A team is testing a new LLM application. During red-teaming, the model consistently leaks sensitive internal project codenames. How should the team address this systematically?
⚠ Common exam trap
Candidates often suggest fine-tuning or retraining the model to remove sensitive data. This is ineffective because models can still hallucinate or reconstruct sensitive information, and it fails to provide a real-time, auditable safety layer.
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
✓
Deploy a guardrail that filters output against a list of sensitive terms.
Systematic mitigation of data leakage requires a multi-layered approach. Modifying the base model is rarely sufficient; instead, one must implement output-side guardrails that perform pattern matching and dictionary-based filtering. This ensures that even if the model attempts to generate sensitive info, the guardrail intercepts it. This process protects intellectual property and maintains compliance with corporate confidentiality agreements by ensuring that protected information remains strictly within the secure environment.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Increase the number of training epochs on the existing dataset.
Why it's wrong here
Increasing training epochs will likely cause the model to memorize sensitive internal codenames even more strongly. Overfitting the model to data that should remain confidential is the opposite of the desired result and will exacerbate the data leakage problem rather than solving it.
- ✓
Deploy a guardrail that filters output against a list of sensitive terms.
Why this is correct
Implementing an output guardrail is the most effective way to intercept sensitive information before it reaches the user. By explicitly defining a list of restricted terms, the system can block or sanitize the response in real-time, providing a robust safety net for protecting internal corporate confidential data.
- ✗
Add a disclaimer at the end of every response stating that the content is confidential.
Why it's wrong here
A disclaimer does not prevent the actual leakage of sensitive information. While legal disclaimers are useful for other purposes, they do not satisfy the security requirement of preventing unauthorized disclosure of confidential data. The data itself must be prevented from being generated or output by the model.
- ✗
Randomize the model's weights during every inference run.
Why it's wrong here
Randomizing weights would degrade the model's performance and factual accuracy, rendering the application useless. It is not a valid strategy for controlling data output. Security must be handled through policy-based guardrails that govern the content of the output, not by introducing instability into the model's mathematical foundations.
About these practice questions
Courseiva writes every NCP-GENL question from scratch — 352 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or 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 NCP-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 NCP-GENL exam.