AI Glossary
Data Poisoning
The manipulation of training data to influence how an AI model learns and behaves.
Data Poisoning
Overview
AI systems learn from data.
The quality of that data often has a significant impact on model performance.
Data poisoning occurs when an attacker intentionally manipulates training data in an effort to influence how a model learns.
A helpful way to think about data poisoning is misinformation during training.
If someone learns from inaccurate or misleading information, their future decisions may be affected.
AI models can experience a similar problem.
By introducing manipulated examples into training datasets, attackers may attempt to alter model behavior, reduce accuracy, or create vulnerabilities.
The effectiveness of data poisoning depends on many factors, including the amount of manipulated data and the training process itself.
Because modern AI systems often rely on large datasets, organizations increasingly pay attention to data quality, governance, and security controls.
Protecting training data helps improve trust, reliability, and model performance.
Why It Matters
Data poisoning can affect how AI systems learn and perform.
Real-World Example
An attacker may attempt to insert misleading examples into a training dataset so that a model learns incorrect patterns.