U.S. Commerce Department Directive DAO 216-26 and the Ban on Differential Privacy

DAO 216-26 Mandates Return to Outdated Privacy Techniques

On June 4, 2026, the U.S. Secretary of Commerce issued directive DAO 216-26, which relegates confidentiality protection for all Bureau of Economic Analysis (BEA) and U.S. Census Bureau publications to statistical techniques from the early 1970s. The directive explicitly bans "noise infusion"—the process of adding random values to a dataset to protect individual identities—and restricts disclosure avoidance to "coarsening" (rounding, aggregating, or using ranges) and "suppression" (redacting values as a last resort).

This policy shift effectively bans differential privacy, the current gold standard for balancing data utility with privacy, as well as other modern techniques like swapping (used since 1990) and input noise infusion (used since 2002). These methods were essential for sharing granular demographic and business data while complying with the Census Act (13 U.S. Code Section 9), which makes it a crime to publish data that allows an individual to be identified.

The Failure of Coarsening to Protect Privacy

Coarsening and suppression are insufficient for protecting granular data because they can be bypassed using basic algebra. When multiple coarsened datasets are released, the intersection of these datasets can allow an attacker to reconstruct exact values, effectively unmasking individuals or businesses.

Case Study: The Brewery Example

To illustrate this vulnerability, researchers provide a scenario involving beer-related businesses in a county with two towns (North Bend and South Bend):

1. The Setup: The county has four entities: a brewery and a bottling company in North Bend, and a brewery and a bottling company in South Bend. Two of these are publicly owned.
2. The Coarsening: The Census Bureau publishes five statistics, coarsening categories to "beer-related" or grouping them by town or ownership to avoid disclosing single-business data.
3. The Result: Despite these good-faith coarsenings, the resulting system of five equations with four unknowns allows anyone with high school algebra to solve for the exact number of employees at each of the four companies.

Noise infusion prevents this exact reconstruction by perturbing the equations, ensuring that individual values cannot be mathematically derived from the aggregate totals.

Political Motivations vs. Scientific Merit

Critics, including Professor Cynthia Dwork and other leaders in computer science, argue that DAO 216-26 is driven by political interests rather than scientific evidence. Specifically, the directive aligns with goals from the Heritage Foundation’s Project 2025 and the Center for Renewing America (CRA).

According to a CRA explainer, the use of differential privacy in the 2020 Census made it "impossible to ascertain the status of individuals" regarding citizenship, even if a citizenship question were added. By banning differential privacy, the administration seeks to make such personal characteristics more accessible, despite the legal requirement to mask this data under the Census Act.

Impact on Data Utility and Public Trust

The ban on modern privacy techniques creates a critical conflict for federal statisticians who must simultaneously provide useful data and maintain legal confidentiality:

• Reduced Data Utility: To avoid legal violations without noise infusion, agencies may be forced to coarsen data so aggressively that it becomes unusable for researchers and policymakers.
• Increased Privacy Risk: Political pressure may lead agencies to publish data that is easily unmasked, violating the law and compromising respondent privacy.
• Erosion of Trust: If respondents believe their data can be easily identified, they are less likely to participate in surveys, leading to a decline in the quality of "democracy’s data."

Community Perspectives and Debate

While the scientific consensus among the authors of the guest post is that noise infusion is necessary, some community discussion highlights a broader debate:

• Implementation Criticism: Some statisticians and political scientists have criticized the implementation of differential privacy in the 2020 Census, arguing it may distort apportionment or redistricting.
• Legal Challenges: There are ongoing lawsuits regarding whether the use of differential privacy in federal statistics is constitutional.
• Systemic Concerns: Commenters on Hacker News expressed skepticism regarding the efficacy of contacting legislators, suggesting that the political system is captured by corporate and political interests, making technical directives like DAO 216-26 a symptom of a larger systemic issue.