AI Glossary

Model poisoning

What does “Model poisoning” mean?

Model poisoning: Deliberately altering training data, updates, or weights to manipulate model behavior.

Model poisoning visual guide: Deliberately altering training data, updates, or weights to manipulate model behavior.
Visual guide Open full size (new tab)

What is an example of “Model poisoning”?

A malicious distributed update introduces behavior the model owner did not intend.

Keep learning

See also

Data privacyTermContent provenance (C2PA)TermResponsible AITermMembership inferenceTermModel extractionTerm

Same topic

Backdoor triggerTermModel monitoringTermOut-of-distribution detectionTermPass@kTermPerplexityTerm