[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"glossary-differential-privacy::en":3,"gloss-cluster-differential-privacy::en":23,"gloss-next-differential-privacy::en":9},{"slug":4,"category":5,"name":6,"definition":7,"meta_desc":8,"faq":9,"schema_markup":9,"related":10},"differential-privacy","security","Differential Privacy","Differential privacy is a mathematical guarantee that the output of an analysis — a statistic, a dashboard metric, a trained model — stays essentially the same whether or not any single person's record is included. It works by injecting carefully calibrated random noise, tuned by a privacy budget (epsilon): smaller epsilon means more noise and stronger privacy, larger epsilon means sharper numbers but weaker protection. The point is that an observer can't reverse-engineer any individual from the results, even with side knowledge. For builders, it's a way to share aggregate analytics, publish benchmarks, or train models on user data while making a real, provable privacy claim rather than a hand-wavy 'anonymized.' Practical note: it's not free — noise degrades accuracy, so it fits aggregate metrics and large datasets far better than small-cohort or per-user precision. Use vetted libraries rather than rolling your own noise, and treat the epsilon budget as a finite resource you spend across queries.","Differential privacy adds calibrated noise so an analysis is essentially unchanged whether or not one person's record is in it — a real mathematical guarantee.",null,[11,14,17,20],{"slug":12,"name":13},"data-residency","Data Residency",{"slug":15,"name":16},"fine-tuning","Fine-Tuning",{"slug":18,"name":19},"model-exfiltration","Model Exfiltration (Model Extraction)",{"slug":21,"name":22},"pii","Personally Identifiable Information (PII)",[24,28,32,36,39,42,45,48,51,54,57,60],{"slug":25,"category":5,"name":26,"updated_at":27},"audit-log","Audit Log (Audit Trail)","2026-08-24T02:46:37+00:00",{"slug":29,"category":5,"name":30,"updated_at":31},"blast-radius","Blast Radius","2026-08-24T03:30:02+00:00",{"slug":33,"category":5,"name":34,"updated_at":35},"break-glass-access","Break-Glass Access","2026-08-24T02:46:38+00:00",{"slug":37,"category":5,"name":38,"updated_at":35},"bridge-letter","Bridge Letter",{"slug":40,"category":5,"name":41,"updated_at":35},"business-associate-agreement","Business Associate Agreement (BAA)",{"slug":43,"category":5,"name":44,"updated_at":27},"byok","Bring Your Own Key (BYOK)",{"slug":46,"category":5,"name":47,"updated_at":35},"cve","CVE (Common Vulnerabilities and Exposures)",{"slug":49,"category":5,"name":50,"updated_at":31},"data-classification","Data Classification",{"slug":52,"category":5,"name":53,"updated_at":35},"data-loss-prevention","Data Loss Prevention (DLP)",{"slug":55,"category":5,"name":56,"updated_at":35},"data-minimization","Data Minimization",{"slug":58,"category":5,"name":59,"updated_at":35},"data-poisoning","Data Poisoning",{"slug":61,"category":5,"name":62,"updated_at":35},"data-processing-agreement","Data Processing Agreement (DPA)"]