3,611 active or planned AI use cases across the federal government — disclosed on 14 April in a one‑line inventory — and a 70% increase from the prior administration’s final catalogue.
Office of Management and Budget inventory: scale, speed, and scant detail
The Office of Management and Budget’s disclosure, published on 14 April, lists 3,611 active or planned AI use cases across federal agencies and marks a roughly 70% increase from the list published in the final year of the Biden administration. The inventory is terse: descriptions are "typically just a sentence, and rarely more than a paragraph," and the public context needed to evaluate purpose, safeguards or risks is largely absent.
Specific agency examples that change how government decides
- The Health and Human Services’ Office of Administration for Children and Families reportedly hired the world’s "scariest AI company," Palantir — "notorious for its work on behalf of the military, the CIA and ICE" — to scan grant applications and flag those "not ideologically aligned with the administration’s dictates."
- The Federal Bureau of Prisons is developing an AI to assess the "potential for misconduct for newly admitted inmates," a system intended to route people into higher security before any misconduct in custody has occurred.
- The Department of Veterans Affairs is developing an AI to listen in on calls to the veterans crisis line, then gather information from external databases to assess a caller’s mental state and suicide risk.
- The Department of Energy is testing AI for model predictive control (MPC) of nuclear reactors as a means to autonomously respond to potential nuclear safety incidents; the disclosure notes that the "recently disclosed addition of AI was initiated under the Biden administration."
- The State Department has ended a program that used AI to forecast mass civilian killings — a program that had been intended to aid conflict prevention.
Translation and routine automation: CBP and the growing plurality of use cases
Not all entries read like dystopian vignettes. Customs and Border Protection has deployed an AI translation system to help officers when human interpreters are unavailable; the inventory lists 70 such translation use cases, up from 58 in the Biden administration’s 2024 disclosure. The essay accepts that machine translation can be preferable to no communication at all, even as it acknowledges the advantages human interpreters bring in nuance and context.
Public consultation, the "high impact" label, and inconsistent guardrails
The inventory’s release is theoretically paired with a consultation process, but in practice the essay reports "there is generally none." Only one of the cited examples — the Department of Justice — "even proposes to involve the public." Under the administration’s policy, broader public engagement is not required unless a use case is classified as "high impact," a label the essay describes as being "applied inconsistently across agencies."
International contrasts: France’s 2016 law and Canada’s 2025 registry
The essay cites two foreign models as instructive. France’s 2016 Digital Republic Act requires algorithms that automate government administrative decisions to be subject to public records requests, appealable to a human reviewer, and to include mandatory notification to affected individuals. Canada, in 2025, launched an AI use case registry and enforces a federal directive mandating transparent risk‑scoring and impact assessment for automated systems that make administrative decisions about citizens; that directive demands a detailed explanation of risks and benefits and consultation with certain stakeholders from conception. The authors note that Canada’s system "could be improved" by adding public comment periods and an obligation for agencies to respond substantively before sensitive uses proceed.
What this means for technologists, policymakers, and the public
- Technologists and security teams: will be asked to build AI systems for roles ranging from translation to nuclear reactor control, but the inventory provides little implementation detail to judge safety, validation or governance requirements.
- Policymakers and regulators: face a choice between piecemeal, internal labelling of "high impact" cases and adopting clearer procedural requirements such as algorithmic impact assessments, public comment and appeal mechanisms — tools referenced in France’s law and Canada’s directive.
- The public and affected individuals: are likely to encounter automated decisions in matters of liberty, health and safety (prison classification, crisis‑line triage, grant awards) even where formal public consultation has not occurred.
The inventory offers a blunt, unavoidable fact: federal agencies are planning and deploying thousands of AI uses that touch migration, criminal custody, mental‑health triage, nuclear safety and grantmaking. The authors argue that many such uses could be pursued responsibly, but only with far greater transparency, consistent risk assessment and genuine public dialogue — practices the current disclosure, as described, does not yet deliver. The essay was written with Nathan E. Sanders and originally appeared in The Guardian.
Original story: https://www.schneier.com/blog/archives/2026/06/ai-use-by-the-us-government.html
