Skip to content

Public-Sector Deployments

Deployment evidence

Public-sector PET deployments can be unusually visible, but they are still contested. Look for both implementation evidence and impact on data users.

Measured Production And Recurring Deployments

2020 U.S. Census Disclosure Avoidance System

Field Entry
Organization / project U.S. Census Bureau 2020 Disclosure Avoidance System
Domain Public sector / official statistics
Problem Release detailed census data while protecting respondent confidentiality against reconstruction and reidentification risks.
PETs used Differential privacy, disclosure avoidance system, post-processing
Deployment maturity Production
Source quality Primary / official plus independent analysis
What worked The Census Bureau used the Disclosure Avoidance System for 2020 Census data products and published extensive documentation.
Challenges The deployment created major utility, communication, and stakeholder-trust debates, especially for small geographies and detailed counts.
Lessons for builders DP deployments need public budget decisions, user education, demonstration data, and a plan for utility disputes. Formal privacy does not remove policy tradeoffs.
Source Census Bureau Decennial Census Disclosure Avoidance and 2020 Disclosure Avoidance FAQ

(Evidence: Deployment-backed / literature-backed. Source quality: Primary / official plus independent analysis. Reviewed 2026-06-17 — production use is clear, and public criticism is part of the evidence base.)

Boston wage-gap analysis using MPC

Field Entry
Organization / project Boston Women's Workforce Council and Boston University Hariri Institute
Domain Public sector / civic labor analytics
Problem Measure gender and racial wage gaps across employers while protecting employer-submitted wage data.
PETs used Secure multiparty computation
Deployment maturity Production, batch / periodic civic analytics workflow
Source quality Primary / official plus third-party case study
What worked BWWC describes using MPC in its wage-gap analysis process and reports aggregated findings.
Challenges MPC protects submitted inputs for the computation, but the released aggregate analysis still needs careful interpretation and cohort controls.
Lessons for builders Civic analytics can use MPC for trust-building, but reporting design and participant communication remain central.
Source Boston Women's Workforce Council: Data Privacy

(Evidence: Deployment-backed. Source quality: Primary / official plus UN case study. Reviewed 2026-06-17 — recurring batch deployment evidence is unusually strong for civic MPC; output privacy remains a separate question.)

Common Proposed Use Cases

Use case Candidate PETs Why proposed Caveats
Official statistics releases DP, synthetic data Agencies must publish useful data under confidentiality mandates Utility loss and public communication are hard
Cross-agency benefit analytics MPC, clean rooms, TEEs Agencies need joint analysis across legal boundaries Governance and purpose limitation may dominate
Public-health dashboards Federated analytics, DP Local data can stay with health departments Small geographies and rare conditions leak
Digital identity and eligibility checks PSI, verifiable credentials, MPC Agencies need yes/no checks without broad data sharing The match result can still be sensitive

Lessons Learned

  • Public deployments need legitimacy, not only technical correctness.
  • Formal privacy parameters become policy choices when outputs affect funding, representation, or rights.
  • Demonstration data and stakeholder review are part of deployment, not afterthoughts.
  • PETs can enable civic collaboration, but they cannot settle acceptable-use disputes by themselves.