Which practice focuses on data validation, cleaning, and anomaly detection to guard against data poisoning?

• A. Data quality in AI security
• B. Explainability in AI
• C. Data validation/cleaning and anomaly detection mechanisms
• D. TEVV in AI system development

Answer: (C)

Data poisoning is mitigated by guarding data quality and integrity through concrete checks on the data itself. The practice that focuses on validating data, cleaning it, and using anomaly detection provides the direct controls needed to identify and remove poisoned or anomalous data before it influences the model. Validating data means ensuring it matches expected formats, ranges, and schemas; cleaning removes suspicious or corrupt samples; anomaly detection flags data points that deviate from normal patterns or historical baselines. Together, these steps create a robust data pipeline that reduces the risk of poisoned data affecting training and inference. The other options touch on related ideas but don’t target the data integrity controls as precisely: data quality in AI security is a broader framing, explainability focuses on interpreting model decisions, and TEVV is about evaluating and validating system performance rather than specifically guarding data against poisoning.