Data Poisoning
An attack where adversarial data is injected into training corpora to corrupt model behaviour. Concern for open-weights models and any model trained on user-contributed data. Detected via provenance tracking and eval-time anomaly detection.
Related terms
- Red-Teaming
Structured adversarial testing of an AI system, either manually or via automated attack pipelines, to find prompt injection, jailbreak, harmful-output, and misuse vulnerabilities before release. Increasingly required by AI regulation for high-risk systems.
- Model Card
A short document describing a model's intended use, training data, evaluation results, known limitations, and ethical considerations. Introduced by Google in 2018 and increasingly required by AI governance regimes (NIST AI RMF, ISO 42001, EU AI Act technical documentation).
Related on this site
Where this fits
Data Poisoning is part of the Evaluation & Governance vocabulary used in the Generative AI Maturity Framework. See the full glossary for the complete set of 149 defined terms, or take the free maturity assessment to see where your organisation stands.