Data poisoning

Topic

Data poisoning is a type of adversarial machine learning attack where an adversary intentionally injects corrupted, false, or misleading samples into a model's training dataset. The goal of this attack is to manipulate the model's learning process so that it behaves incorrectly, makes inaccurate predictions, or exhibits specific backdoors during deployment. This technique poses a significant security risk to artificial intelligence systems, particularly those that continuously learn from user-generated or web-scale data.