AAISM: Advanced in AI Security Management

Data poisoning is best detected by monitoring the training dataset for unusual or anomalous patterns, because malicious or manipulated records are introduced during data collection or preparation and may otherwise influence learned model behavior.