US Government Discloses 3,611 Uses of AI in Federal Agencies

The Trump administration has disclosed data on large-scale use of artificial intelligence in federal government bodies: 3611 active or planned use cases, 70% more than under the previous administration.

A quiet admission

On April 14, the Office of Management and Budget (OMB) published an updated inventory of AI use across US federal agencies. Notably, it happened without press conferences or public hearings: the data appeared as a technical commit in a government GitHub repository.

The list grew 70% compared with the equivalent inventory from the final year of Biden's presidency and covers thousands of scenarios in which algorithms make or support decisions on behalf of the state. For scale: just a few years ago there were hundreds of pilot projects, mostly in document management. Today the figure has passed three and a half thousand.

Some of these are relatively harmless support tasks: automating request processing, searching databases, classifying documents. But a significant share of the inventory concerns decisions that directly affect the lives and freedom of specific people.

What is moving under algorithmic control

The inventory covers a broad and, in places, troubling range of sensitive government functions. Tasks delegated to AI include:

• Risk assessment systems in the justice system, which influence decisions on arrest, bail and parole
• Management of access to government healthcare and social security services
• Safety monitoring of nuclear reactors and critical infrastructure
• Border control and immigration decisions
• Predictive analytics in law enforcement

These are not abstract prospects: the systems are already being deployed or are scheduled for launch at all levels of federal government.

Transparency that isn't there

Nathan E. Sanders of Harvard's Berkman Klein Center and Bruce Schneier of the Kennedy School of Government, authors of the book Rewiring Democracy: How AI Will Transform Our Politics, Government, and Citizenship, put the core problem this way:

"This is a handover of decision-making processes from humans to

machines on an enormous scale, on matters of personal liberty, people's health and well-being, the safety of nuclear facilities and much more."

The key issue is not the use of AI itself, but the lack of accountability and appeal mechanisms. When an algorithm rules on immigration status or on the risk of releasing a prisoner, who is responsible for an error? How do you challenge a decision when the criteria behind it are not disclosed? Many government algorithms work as "black boxes": their internal logic is closed even to regulators.

What this means

Disclosing 3611 cases is a step toward transparency, but the way it happened is telling. The public learned of it not through congressional hearings or official statements, but through a technical commit on GitHub. When a large-scale transfer of government functions to algorithms happens without public discussion, the risk of systemic errors grows while the tools to fix them shrink.

The question is no longer whether AI will run critical functions of the state. It already does. The question is who controls the AI itself.