Question 28
Domain 5: GovernanceA team discovers that a data source feeding its GenAI application contains problematic text that cannot be reliably cleaned before ingestion. What is the best mitigation approach?
Correct answer: C
Explanation
When problematic text in a source cannot be effectively remediated, the safest alternative is to replace or exclude that source rather than pass the risk downstream into the GenAI system. — official.txt
Why each option is right or wrong
A. Keep the data source and rely on the model to ignore problematic text during generation
Problematic text should be mitigated in the data source feeding the GenAI application, not left for the model to handle later.
B. Increase the amount of problematic text so the model learns that it is common input
Mitigation reduces problematic text exposure; adding more problematic text increases risk in the data source.
C. Use an alternative data source that does not contain the problematic text
The source material directs recommending an alternative for problematic text mitigation in a data source feeding a GenAI application. Because this source cannot be reliably cleaned before ingestion, replacing it with an alternative data source is the appropriate mitigation approach.
D. Delay mitigation until after deployment and address issues only if users report them
Problematic text mitigation applies to the data source feeding the GenAI application before downstream use.