Definition
Model poisoning is a type of attack against artificial intelligence systems where malicious actors intentionally corrupt the training data to degrade the model's performance or cause it to make specific incorrect predictions. This manipulation can occur during the data collection or labeling phases, subtly altering the patterns the AI learns. Understanding model poisoning is crucial for maintaining the integrity and reliability of AI applications in production environments. For instance, an attacker might inject mislabeled images into a facial recognition training set to cause misidentification. This differs from adversarial examples, which target a trained model rather than its training data. Concepts related to data integrity and security are frequently assessed in cybersecurity certifications like CompTIA Security+ and CySA+.