Dictionary entry
data poisoning
also training-data poisoning or poisoning attack
Definition of data poisoning
-
: an attack that inserts, changes, or selectively removes examples in a machine-learning system's training data in order to degrade the model, bias its behavior, or create a hidden response to a trigger
- The attacker attempted data poisoning by adding mislabeled samples to the public training set.
- A few carefully crafted documents created a backdoor that activated only when the model saw a particular phrase.
Two meanings, one word
In everyday English
Poisoning means contaminating something so that it becomes harmful or unsafe.
In AI
An attack that inserts, changes, or selectively removes examples in a machine-learning system's training data in order to degrade the model, bias its behavior, or create…
Origin & history
The term grew out of computer-security poisoning attacks and became standard in adversarial machine learning as researchers studied attackers who could influence part of a model's training data.
Test yourself
Which of these is the meaning of data poisoning?
Cite this entry
"Data poisoning." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/data-poisoning/. Accessed 7 Oct. 2026.