What is Data Poisoning?
Injecting malicious data into model training data, making the model learn wrong patterns or backdoors. Like sneaking incorrect knowledge into textbooks. The attack surface is wide β from public datasets to fine-tuning data, all can be poisoned.