The Government Run by Algorithms

Artificial intelligence is quietly moving into the machinery of American government. The question is no longer whether agencies will use it—but how much authority Americans are willing to give it.

WASHINGTON — For generations, Americans dealing with the government encountered clerks, investigators, police officers, auditors and bureaucrats. Increasingly, another participant is entering those interactions: an algorithm.

Artificial intelligence is spreading through government at remarkable speed. Federal agencies are experimenting with AI to detect fraud, select tax returns for examination, analyze benefit claims, identify travelers, assist immigration operations and improve public services. State and local governments are deploying related technologies for policing, traffic management and administrative work.

The transformation is already significant. The Government Accountability Office found that across 11 federal agencies it examined, reported AI applications nearly doubled from 571 in 2023 to 1,110 in 2024. Generative-AI use increased ninefold during roughly the same period.

What sounds like an information-technology upgrade could eventually become something much larger: a fundamental change in how government exercises power.

The Algorithm at the IRS

Consider the Internal Revenue Service.

According to a March 2026 GAO report, the IRS had 126 active AI applications in its inventory as of June 2025, compared with just 10 reported applications in August 2022. The technology is being used in areas including audit selection, taxpayer assistance, fraud detection and tax compliance.

The attraction is obvious.

The IRS receives enormous quantities of financial information. An algorithm capable of examining millions of transactions can identify patterns that would take human investigators years to discover.

That could mean catching sophisticated tax fraud while reducing unnecessary audits of legitimate taxpayers.

But it also raises a difficult question: What happens when the algorithm is wrong?

A taxpayer targeted by an automated system may never know exactly why the computer considered his return suspicious. And sophisticated AI models can become difficult even for their operators to fully explain.

AI at the Border

The stakes become greater when algorithms move from paperwork to decisions involving people's identities and movements.

The Department of Homeland Security has experimented extensively with AI. Its applications have included facial recognition, traveler verification, port-of-entry risk assessments, identity matching, geospatial disaster assessments and immigration-related systems.

The federal government itself recognizes that certain applications deserve greater scrutiny. Current DHS guidance categorizes areas including law-enforcement risk assessments, biometric identification, crime forecasting, immigration risk assessments, benefits adjudication and fraud detection as potentially “high-impact AI.”

The reason is simple: these systems can influence decisions with serious consequences for individuals.

An algorithm recommending where police should patrol is considerably different from software organizing government emails.

The Promise

There are compelling arguments for expanding government AI.

Government agencies possess mountains of information but limited manpower. AI can potentially identify Medicare fraud, detect suspicious financial activity, process routine applications, translate documents and help overwhelmed employees locate important information.

The technology could also make government less frustrating.

Instead of waiting hours for assistance, a citizen might receive immediate answers from an AI system. Veterans' claims could be reviewed faster. Disaster agencies could analyze aerial imagery within minutes. Investigators could identify patterns across millions of records.

GAO has documented federal uses ranging from facial recognition at airports to analysis of veterans' benefit claims.

Used correctly, AI could make government faster, cheaper and more effective. But efficiency is not the same thing as justice.

When the Computer Says No

The danger becomes clearest when algorithms move from assisting humans to effectively replacing human judgment.

Imagine an algorithm determining that someone is likely committing benefits fraud.

Or identifying someone as a security risk.

Or influencing whether a neighborhood receives additional police surveillance.

Computers can analyze information without fatigue, emotion or favoritism. But algorithms learn from data created by humans and institutions. If historical data contains distortions or biases, automated systems can reproduce them—potentially at enormous scale.

There is another problem: accountability.

A government employee can theoretically explain a decision. A supervisor can review it. A citizen can appeal it.

But who is responsible when an AI system recommends the wrong person?

The programmer?

The private contractor?

The government agency?

Or the official who followed the computer's recommendation?

Those questions become particularly important because federal agencies frequently acquire AI capabilities from private companies. GAO has warned that agencies face difficulties finding enough technical expertise to evaluate AI products and even understanding their costs.

A Government Transformation Already Underway

Washington has begun constructing oversight mechanisms. GAO identified numerous government-wide AI requirements involving inventories, management, transparency and oversight, while agencies are required to publicly document many non-sensitive AI applications.

Yet regulation is chasing technology that evolves extraordinarily quickly.

And that may ultimately produce the defining government debate of the AI era.

The issue is not whether government should use artificial intelligence. That decision has largely been made.

The more important question is where Americans draw the line.

Few people will object to an algorithm helping a clerk locate a document or helping FEMA analyze hurricane damage. The debate becomes considerably different when software influences whether someone is investigated, audited, denied government assistance, stopped at a border or identified by police.

The challenge for government, therefore, may not be building smarter machines.

It will be deciding which decisions machines should never be allowed to make alone.

Because the most consequential question surrounding artificial intelligence in government isn't whether an algorithm can make a decision.

It is whether it should.