// Hacker Noon · 26 March 2026

Model Poisoning Turns Helpful AI Into a Trojan Horse

Model poisoning is the malicious manipulation of a machine learning model's training data or parameters to embed hidden, "backdoor" behaviors. The attack works in four steps: Poisoning the weights, triggering triggers, exfiltrating data, and hiding the data.

Hacker Noon

@hacker-noon · Felix Koole

hackernoon.com

Hacker Noon@hacker-noon