A model is (ε, δ)-differentially private if its output distribution changes negligibly when any single training record is added or removed. Useful for training on sensitive data — but the noise costs accuracy, and the ε budget is easy to burn through.
GlossaryPart of the BotDoors AI Safety & Compliance Glossary · 37 terms.
Security
Differential Privacy
Differential privacy adds mathematically calibrated noise so no individual record can be reverse-engineered from a model.