The BEST way to prevent sensitive information disclosure by large language model (LLM) chatbots is through:

The correct answer is D. Data masking. Data masking replaces sensitive values (PII, credentials, financial data) with realistic but fictitious substitutes in training datasets and pipeline inputs, ensuring the LLM never ingests or memorizes actual sensitive information and therefore cannot disclose it. This is a…

AI Risk Management and Controls

Question

Options

AManual monitoring
BAccess controls
CData sanitization
DData masking

How the community answered

(58 responses)

A
14% (8)
B
22% (13)
C
7% (4)
D
57% (33)

Explanation

Data masking replaces sensitive values (PII, credentials, financial data) with realistic but fictitious substitutes in training datasets and pipeline inputs, ensuring the LLM never ingests or memorizes actual sensitive information and therefore cannot disclose it. This is a proactive, preventive control applied at the data level. Data sanitization (C) removes or scrubs data but is less precise and may not cover all sensitive patterns within otherwise valid training records. Access controls (B) limit who queries the model but do not prevent the model itself from reproducing sensitive content it has learned. Manual monitoring (A) is reactive and cannot scale to the volume of LLM outputs.

Topics

#LLM security#Data privacy#Data masking#Sensitive information disclosure

Community Discussion

No community discussion yet for this question.

Full AAIA Practice