As of the 2025-2026 source window, what were the two leading practical applications of differential privacy?

Generating synthetic datasets that can be shared freely, and answering well-defined repetitive queries where the privacy budget and noise are well understood.

As of 2025-2026, DP could not yet power an arbitrary-query private server, but two uses worked well in practice:

• Synthetic data generation — use DP to generate an artificial dataset that preserves statistical properties while protecting individuals; the synthetic data can then be shared more freely.
• Simple repetitive queries — DP is successfully deployed for well-defined, repetitive query patterns (e.g. daily telemetry counts) where the privacy budget can be carefully managed and the noise calibration is well understood.

Both sidestep DP's weaknesses (unbounded queries, budget exhaustion) by constraining the query space.

Tip: The pattern is "constrain the questions, then DP becomes practical." Synthetic data does this by answering all questions once, up front.