LLM04

OWASP Top 10 for LLMs / Data and Model Poisoning

Attackers corrupt the data used to train or fine-tune a model, planting biases, backdoors, or flaws that quietly shape how the model behaves later.

Think of it likeSomeone slipping bad ingredients into the flour bin so every loaf your family bakes for months comes out tainted.

In plain English

If the examples an AI learns from are secretly rigged, the finished model carries hidden flaws or backdoors that show up long after training.