DAO 216-26 Orders Census to Use 1970s Privacy Tech, Ignoring Fifty Years of Progress

On June 4, 2026, the U.S. Secretary of Commerce issued DAO 216-26, a directive that effectively bans differential privacy and all noise-infusion techniques from Census Bureau and BEA publications, mandating only coarsening and suppression—methods that proved insufficient for granular data decades ago. Cynthia Dwork, Gordon McKay Professor of Computer Science at Harvard, and a coalition of leading theorists have posted a stark warning on Scott Aaronson's blog: this order throws away over half a century of progress in disclosure avoidance, leaving both privacy and data utility worse off.

How DAO 216-26 Reverses 50 Years of Disclosure Avoidance

The directive restricts acceptable techniques to “coarsening” (rounding, aggregating, using ranges) and “suppression” (redacting specific values) as a last resort. It explicitly forbids “noise infusion”—adding random values to data—which includes input noise infusion (used in Quarterly Workforce Indicators since 2002), swapping (used in decennial census publications since 1990), and differential privacy (used for OnTheMap since 2008 and for 2020 Census publications). Differential privacy was slated for the 2030 Census; that plan is now dead.

The Real Target: Preventing Identification of Citizenship Status

The acting force is political, not scientific. The directive bypasses standard administrative procedures and fulfills a promise from the Heritage Foundation’s Project 2025. The Center for Renewing America (CRA), founded by OMB Director Russell Vought, published an explainer that makes the motive explicit: “Even if the citizenship question is added to the Census, it will be impossible to ascertain the status of individuals so long as differential privacy is used.” The problem? The Census Act (13 U.S. Code Section 9) makes it a crime to publish data from which any particular individual can be identified. Differential privacy is the best tool we have for satisfying that mandate while releasing useful granular statistics.

Why Coarsening and Suppression Can't Handle Fine-Grained Data

Nathan Goldschlag’s brewery example from County Business Patterns illustrates the tension. One brewery in a small county? Publishing exact employee counts identifies that business—a legal violation. Two breweries? A competitor can deduce the other’s headcount. More than two? Coarsening might suppress the cell entirely or aggregate it to a meaningless level. Noise infusion handles these cases by adding calibrated randomness, preserving utility while meeting confidentiality requirements. DAO 216-26 eliminates that option, forcing the Census Bureau to either publish less data (reducing its fitness for use) or publish data that risks reidentification (violating the law). The directive is incompatible with the bureau’s dual mandate.

If DAO 216-26 stands, the 2030 Census will be the least useful and least protective in modern history—precisely when granular economic and demographic data is most needed to inform policy and research.

Source: An American Privacy Emergency
Domain: scottaaronson.blog