Concept dictionary
Model Poisoning
Model poisoning is a deliberate attack that corrupts training data, learning processes, model updates, or model artifacts to alter behavior or introduce hidden weaknesses.
- Review state
- Needs review
- Freshness
- Current
- Version
- 2
Also called
- Model Poisoning
- Data Poisoning
Remember it this wayPoisoning changes what the model learned or carries.
In plain language
The attacker interferes before or during model development rather than only at the moment a user sends a request. The effect may be broad degradation or a targeted behavior triggered under specific conditions.
Why it matters
A compromised model can pass ordinary tests while retaining a hidden failure path. Supply-chain controls, provenance, access restrictions, and targeted evaluation matter.
Example
An attacker inserts manipulated examples into a training pipeline so a particular visual pattern causes the model to assign a chosen label.
Common misconception
Model poisoning is different from prompt injection. Poisoning alters training or model integrity; injection attacks a deployed system through processed instructions.
Related terms
Sources
- NIST AI 100-2 E2025, Adversarial Machine Learningauthoritative taxonomy of poisoning attacks and mitigations.
Learn more
- Use [[data-provenance|Data Provenance]] to examine the training supply chain.