Synthetic Data

Synthetic data is artificially generated data used for training, testing, or privacy-preserving sharing when real data is scarce or sensitive.

Made-up-but-realistic data created by software to train or test systems.

Synthetic data is generated by models or simulators for training, testing, or privacy-preserving sharing. Utility and leakage risks must be evaluated—not assumed.

Teams use synthetic data when real data is scarce, sensitive, or imbalanced—but poor synthetics can mislead models.

• Augmenting rare classes
• Privacy-preserving sharing
• Simulation for robotics/autonomy

• Can fill data gaps
• May reduce exposure of raw PII when done carefully

• Distribution mismatch
• Potential privacy leakage from generative models

• Synthetic does not automatically mean anonymized

Is synthetic data always safe to share?

No. Evaluate re-identification and membership risks; use privacy techniques when required.

• NIST AI RMF

• Synthetic data privacy and utility literature