Data Poisoning is defined as?

• A. Manipulating Training Data To Cause Biased, Inaccurate, Or Malicious Outputs
• B. Data Encryption To Prevent Leakage
• C. Normalizing Data Before Training
• D. Removing Outliers From Dataset

Answer: (A)

Data poisoning is about compromising the training data to steer the model’s learning toward biased or harmful results. Attackers may flip labels, insert mislabeled or crafted instances, or embed backdoors that trigger specific outputs when a hidden input pattern appears. The result is biased, inaccurate, or malicious outputs, eroding trust and model performance. By contrast, data encryption protects confidentiality, while normalization and removing outliers are standard preprocessing steps that aim to improve accuracy and consistency, not corrupt the model’s behavior.