AI use by the US government is ballooning. And the lack of transparency is troubling | Nathan E Sanders and Bruce Schneier
TL;DR
OMB lists 3,611 active or planned AI use cases across US federal agencies. Sanders and Schneier say the inventory is up 70 percent from the last Biden-era disclosure. Examples include Palantir screening HHS grant applications, Federal Bureau of Prisons risk scoring for new inmates, and AI analysis of Veterans Crisis Line calls. The Department of Energy is testing AI responses for nuclear reactor safety incidents, while the State Department has ended an AI project meant to forecast mass civilian killings.
Nauti's Take
This is the exact zone where AI optimism becomes lazy. Translation support and internal research can make government more accessible, but risk scoring people or automating responses in critical infrastructure needs more than a GitHub inventory.
The US approach described here looks less like disciplined modernization and more like automate first, justify later. Good public-sector AI starts with rights, appeal routes and public accountability, not model deployment counts.
Briefingshow
Government AI changes who effectively prepares state decisions: not only civil servants, but models, data pipelines and classification rules. For prison security, suicide risk or nuclear safety, one-line disclosures are not enough because errors can hit rights, safety and trust directly. The article's strongest point is procedural: transparency without risk assessment, human appeal and public input is mostly a spreadsheet exercise.
Sources