Data poisoning occurs when attackers manipulate training data to corrupt model behavior. The most direct mitigation is to implement robust data validation and integrity checks (option C), including anomaly detection on input distributions, provenance controls, verification of data sources, and safeguards for pipelines feeding the training set. AAIA highlights threats and vulnerabilities specific to AI and the importance of controls that protect data integrity in AI Operations.
Option A (relying on third-party providers) does not inherently eliminate poisoning risk; providers themselves may be vulnerable. Option B (increasing data size) can dilute but not reliably remove malicious samples.
Option D (simpler algorithms) might help interpretability but does not directly prevent poisoned data from influencing the model. The most effective, aligned control is rigorous data validation to ensure only trustworthy data enters the training process.
References:
ISACA, AAIA Exam Content Outline - Domain 2: Threats and Vulnerabilities Specific to AI (controls for AI- related threats).
ISACA risk guidance referencing data integrity and poisoning risks in AI pipelines.