Who uses differential privacy in production, and how do Apple's and Google's approaches differ?

Apple, Google, the US Census Bureau, and Microsoft all deploy DP; Apple uses on-device (local) DP and keeps it closed-source, while Google open-sourced its DP library.

The headline split is where the noise is added, and that drives almost everything else about each deployment. Apple does it on your phone, so it never sees your raw answer; Google does it more centrally but gives the tooling away. Both prove DP has moved from theory into shipping products, though each had to wrestle with the privacy-versus-utility balance.

• Apple uses local differential privacy — noise is added on-device before anything is sent — and has done so since iOS 10 / macOS Sierra, one of the first at-scale adoptions. It collects things like which sites drain battery (power usage), emoji-in-context, and new words people type that aren't yet in the dictionary. Crucially, the implementation is proprietary and closed-source, so outsiders can't audit the exact guarantee.
• Google took the opposite stance and open-sourced its DP library, making the technique reusable; it deploys DP in Chrome (browser-malware statistics) and Maps (aggregate traffic data).
• The US Census Bureau applied DP to the 2020 Census, a landmark government adoption that protected respondents while publishing usable population tables.
• Microsoft folds DP into product telemetry and diagnostics.

Tip: "Local" DP (Apple) protects you before data leaves your device — you needn't trust the collector. "Central" DP trusts a curator to hold raw data and add noise to query results — easier to build, but you must trust the curator.