A US Department of Commerce order barring “noise infusion” from statistical products published by the Census Bureau and Bureau of Economic Analysis has drawn criticism from privacy researcher Damien Desfontaines. He argues that the directive removes important tools without resolving agencies’ legal duty to protect confidential records.
Public statistics are commonly calculated from data that cannot be released person by person. Disclosure-avoidance methods are intended to let agencies publish useful totals while reducing the chance that an attacker can reconstruct an individual record. Desfontaines describes techniques including swapping, limiting contributions, suppression and coarsening. Differential privacy typically combines carefully calibrated random noise with other controls to quantify the relationship between privacy risk and accuracy.
According to the account, the Census Bureau mainly used swapping for decennial census releases from 1990 through 2010. The agency later found that published tables could be used to reconstruct records and selected differential privacy for the 2020 census after comparing alternatives. Desfontaines says that choice preserved more utility than other approaches capable of addressing the reconstruction risk, although it still made some figures less accurate than their 2010 counterparts.
The new order favors coarsening and describes suppression as a last resort, while stating that it does not override constitutional, statutory or regulatory requirements. That qualification is central to the critique: confidentiality obligations remain, but the agencies have fewer ways to satisfy them while producing detailed data about small populations.
Desfontaines warns that exact-looking figures can make reconstruction easier because attackers can treat published values as fixed equations. Randomness forces an adversary to account for uncertainty. He also notes that alternatives are not always free of random effects. Swapping changes which records contribute to a local result, while sampling and imputation can also produce estimates that differ from underlying population values. Whether every such practice falls within the order’s definition is not clear from the supplied account.
The commentary predicts that agencies will face a harsher choice: publish less detailed information, accept weaker privacy protection, or combine both outcomes. It particularly questions whether blunt suppression and aggregation can preserve information about small minority groups.
This is an expert interpretation of the order, not an agency impact assessment, and the supplied evidence does not include the directive itself or a Commerce Department response. The immediate verified development is therefore the reported restriction and a technical warning about its consequences. Implementation guidance, definitions and the agencies’ eventual publication methods will determine how much the policy changes specific statistical products. Those choices may also affect public confidence in official data.
