On April 14, the Office of Management and Budget disclosed 3,611 active or planned artificial intelligence use cases across the federal government - a 70% jump from the final Biden-era inventory. The list, quietly published on the OMB's federal chief information officer's GitHub page and noticed mainly by outlets like FedScoop, shows agencies rapidly shifting sensitive functions to automated systems, often with almost no public oversight.
The Health and Human Services office that administers children and family programs has deployed Palantir software to scan all grant applications and flag those that don't align with administration policy. The Federal Bureau of Prisons is building an AI system to assess "potential for misconduct for newly admitted inmates" and route people into high-security confinement before any infraction occurs inside the facility. The Department of Veterans Affairs is testing AI that listens to calls on the veterans crisis line, then pulls data from external databases to assess suicide risk.
Details sparse in AI inventory
The Department of Energy is testing AI for autonomous control of nuclear reactors to respond to safety incidents. The State Department retired a program that forecast mass civilian killings, an initiative that had been designed to aid conflict prevention. While any of these programs could be implemented with rigorous safeguards, the inventory descriptions are typically a single sentence and rarely explain how the AI actually works or what human oversight remains.
Under the administration's policy, most of these use cases are not classified as "high impact," so they require no public consultation. Only one - a Department of Justice initiative - proposes any form of public involvement. The gap between the inventory's breadth and its disclosure detail makes it nearly impossible for citizens or oversight bodies to assess risks.
Government professionals working with AI tools face a similar problem. Without transparent frameworks, even well-intentioned automation can erode public trust. Training resources like AI for Government can help teams understand how to deploy these systems responsibly while meeting compliance and ethical standards.
International models for algorithmic transparency
Other governments offer a path forward. Canada's 2025 AI use case registry is backed by a federal directive that mandates a transparent risk-scoring and impact assessment process for automated systems that make administrative decisions about citizens. That directive requires a detailed explanation of risks and benefits, as well as stakeholder consultation from the earliest stages of planning. France's 2016 Digital Republic Act requires all algorithms used in administrative decisions to be subject to public records requests, appealable to a human reviewer, and accompanied by mandatory notification when automation is used.
Washington, D.C., and California have also held large-scale public deliberations on government AI, using online platforms to gather broad input. These examples show that disclosure paired with meaningful public engagement can build confidence rather than suspicion. For policy makers shaping these rules, the AI Learning Path for Policy Makers provides a structured way to evaluate risks, design governance models, and align automation with democratic values.
Why this matters for government professionals
The explosion of federal AI use cases will inevitably reach state and local agencies, law enforcement, health systems, and infrastructure operators. Public servants who work with these systems need to insist on clear documentation, defined human review points, and a credible process for affected people to challenge automated decisions - well before the system is live. Canada's mandatory impact assessments and France's right to human review are concrete models that can be adapted in the U.S., and they highlight that transparency is not a bureaucratic box to check but a prerequisite for safe and legitimate government use of AI.
Your membership also unlocks:
