Data Poisoning is best defined as?

• A. Manipulating Training Data To Cause Biased, Inaccurate, Or Malicious Outputs
• B. Data Encryption To Protect Secrets
• C. Running Tests On Data
• D. Archiving Data After Training

Answer: (A)

Data poisoning involves manipulating the training data used to build a model with the aim of causing biased, inaccurate, or malicious outputs when the model is deployed. This type of attack tamperes with the information the model learns from, such as injecting poisoned samples, flipping labels, or subtly shifting data distributions, sometimes even creating a hidden trigger (a backdoor) that activates under specific conditions. Because it targets the data the model relies on during learning, it directly changes behavior rather than protecting data, evaluating it, or archiving it. The other options describe encryption, evaluation, or archiving, which do not address altering model behavior through training data.