Generative AI Feature Additions Increase Risk by Expanding Attack Surface
Short answer
B Adding features to the generative AI product increases risk because each new function expands the attack surface and creates additional ways for attackers to interact with the model.
Question
Key Takeaways
This question tests whether you link AI risk to exposed product functionality rather than only to model capability; the trap is treating a more capable model as automatically riskier.
In a generative AI product, adding functionality expands the attack surface and generally increases risk. This page establishes why option B is the best answer and why model capability and qualitative measurement do not settle the question.
Common mistake: The most plausible wrong answer is D: risk is not proportional to model capabilities in this scenario, because a capable model can be tightly controlled while a modest model exposed through an insecure new feature can create greater risk.
Expert Analysis
Risk is driven by exposed functionality The scenario is about a generative AI product adding features to compete, not about changing the underlying model's training or parameter count. Each new customer-facing capability—whether a plugin, file-upload path, data connector, autonomous action, or new API endpoint—gives users and attackers another way to interact with the system. That expands the attack surface, so the most accurate statement is B: risk increases when new features are added. The supplied community explanation makes the same point, noting that new features create additional ways for attackers to interact with the model.
The tempting boundary: capability versus functionality Option D sounds plausible because AI risk discussions often mention model capability. But the question's boundary is between a model's intrinsic capability and the product's enabled functionality. A highly capable model can be deployed with strict permissions, limited data access, and strong monitoring; a less capable model can be exposed through an insecure feature that accepts untrusted input or connects to sensitive data. Capability contributes to risk, but it does not define the direction of risk when functionality is enabled.
Measurement is not direction Option C confuses how risk is assessed with whether it rises. Qualitative measurement can be appropriate, but it does not answer what happens when functionality expands. Option A is too absolute: adding features does not necessarily change every individual risk, but it can introduce new threats and failure modes, so the overall risk posture generally moves upward.
Verdict B is the best answer: new features expand the generative AI product's attack surface and increase risk, even though the exact increase depends on controls and data exposure.
Official Reference
- https://www.nist.gov/itl/ai-risk-management-framework
- https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf
- https://owasp.org/www-project-top-10-for-large-language-model-applications/
These links open the official vendor documentation in a new tab.
Exam Strategy
When a question asks how risk changes as AI functionality is added, trace each new feature to a concrete exposure: endpoint, plugin, data connection, permission, or user input path. Eliminate answers that describe how risk is measured (C) or tie risk only to model capability (D), because neither answers the direction of change. Pick the option that says risk rises when the attack surface grows.
Frequently Asked Questions
Does adding a feature increase risk even if the underlying model is unchanged?
Why is D wrong—aren't more capable models riskier?
Is risk always proportional to the number of features?
Discussion 1 contributions
Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.
From public community threads 1
-
Comptia SecAI Group 👍 1 Answered BThe risks increase, as the attack surface expands with new features and potential ways for attackers to interact with the model.
Reader comments & corrections
No reader corrections yet — spotted an error or have a note? Share it below.