A type of adversarial attack where malicious data is intentionally introduced into a machine learning model's training dataset to corrupt its behavior and decision-making capabilities.
Model poisoning represents an insider threat to AI systems where employees with access to training data or model development can introduce biased or malicious samples to compromise model integrity. This can affect insider threat detection systems themselves, causing them to ignore certain attack patterns or generate false alerts. Defense strategies include data validation, robust training techniques, and strict access controls over training datasets and model development environments.