What is AI poisoning? Definition and how to prevent it

AI poisoning is a deliberate attempt to manipulate what an AI system learns, remembers, or relies on, so that it produces false, biased, or malicious outputs later. Unlike traditional hacking, which often exploits software bugs, AI poisoning targets the data and context the model relies on. An attacker tries to influence what the model treats as true, safe, or normal.