AI Is Running Government. Accountability Is Not Keeping Up.

AI Is Already Running Your Government. Most Agencies Do Not Know How Much.

Most conversations about AI in government focus on the debate: should agencies be using it, how fast should they move, and who gets to decide. That debate is already settled by the data. In 2025, 41 federal agencies documented more than 3,600 individual AI use cases, 69% above the total reported in 2024 and five times the number reported in 2023 (Brookings, April 2026). AI is not coming to the government. It is already there, already running, already influencing decisions that affect millions of people every single day. The question that actually matters is whether the structures built to manage it are keeping pace with the speed at which it is being deployed. Right now, the honest answer is no. And the gap between deployment and accountability in the public sector is not just a technology problem. It is a public trust problem.

The Gap the Market Is Misreading

Most people think of government AI as a future-state conversation. Something being planned, piloted, and slowly rolled out through carefully managed procurement cycles. That picture is about two years out of date. Federal AI spending increased by 966% between 2024 and 2026, from 355 million dollars to 7.2 billion dollars obligated. The number of federal agencies with AI contracts rose from 17 in 2022 to 28 in 2026, with the total number of contracts growing from 472 to 1,743 over the same period (Brookings, June 2026).

That is not a slow, deliberate rollout. That is an institution moving fast. And at the state and local level the picture is just as significant. 55.7% of government organizations now use AI, but only 42.9% report having formal AI policies in place. Operationalization is outpacing standardization across state and local agencies, with digital form responses jumping nearly 30% between 2024 and 2025 as AI becomes a force multiplier for agencies managing growth without proportional headcount increases (Granicus, April 2026).

What Government AI Actually Looks Like in Practice

Government AI is not one thing. It shows up differently depending on the agency, the function, and the level of government involved. Understanding what it actually looks like in practice is the first step toward understanding where the accountability gaps are most significant.

At the federal level, AI is embedded in the decisions that affect daily life. The Social Security Administration uses AI to support 52% of its service delivery and benefits processing use cases. The Department of Veterans Affairs uses AI in 45% of its health and medical services use cases. The Department of Justice uses AI in 54% of its law enforcement use cases (Brookings, April 2026). These are not administrative back-office functions. These are the systems that determine whether a veteran receives healthcare benefits, whether a Social Security claim is processed correctly, and whether a law enforcement decision is made with accurate information. The stakes attached to those decisions are not abstract.

At the state and local level, AI is being used to manage the growing volume of public services. State and local agencies are moving beyond pilots and embedding AI directly into investigative casework, eligibility enforcement, provider oversight, and fraud investigation. Agencies responsible for licensing and eligibility report staffing constraints as their primary risk as caseloads rise without headcount growth, and AI is being used to automate repeatable tasks so staff can focus on critical case functions (Route Fifty, April 2026). That is a practical and understandable response to a real operational challenge. It is also a deployment pattern that requires careful governance around how automated decisions get made and how they get reviewed.

Generative AI specifically is growing faster than any other category. From 2023 to 2024, federal agencies' use of generative AI increased ninefold. Across 11 selected agencies, the total number of reported AI use cases nearly doubled from 571 in 2023 to 1,110 in 2024, while agencies report significant challenges complying with federal policies while keeping up with rapidly evolving technology (GAO, July 2025). Ninefold growth in a single year means governance frameworks that were adequate in 2023 were already underpowered by 2024. The agencies that have not updated their accountability structures since then are running a significantly different set of AI tools under a significantly older set of rules.

Why the Accountability Gap Creates Disproportionate Public Risk

The cost of an accountability gap in government AI is not the same as the cost of one in a private enterprise. When a private company's AI produces a biased or inaccurate output, it affects a customer relationship. When a government agency's AI does the same thing, it affects a citizen's access to benefits, legal standing, or public services. The stakes are higher, the populations affected are more vulnerable, and the consequences are harder to reverse.

The transparency mechanisms designed to catch problems are not working as intended. Federal agencies reported 3,611 AI use cases in 2025, a nearly 70% increase from 2024. But the Department of Justice did not include any risk management information for its use cases for the second year in a row, despite 114 of its 315 use cases being deemed high-impact. The Department of Homeland Security reported a number of significantly risky use cases but determined that a high percentage did not qualify as high-impact and were therefore not subject to heightened risk management requirements (CDT, April 2026). Transparency mechanisms that allow agencies to self-classify their own risk levels are not independent accountability structures. They are documentation exercises. And documentation exercises do not catch the outcomes that matter most.

Bias in government AI is already producing documented inequitable outcomes. The IRS was auditing Black taxpayers at three times the rate of other taxpayers based on outputs from an AI system as recently as 2023, illustrating how government AI can perpetuate human biases contrary to the commonly held perception that algorithmic outputs are strictly objective (AAAS, 2025). This is the same dynamic we covered in the healthcare and hospital AI briefs. The data that trains a government AI system reflects the inequities already present in the system it was built from. Without deliberate design and continuous monitoring, those inequities do not disappear. They get automated.

The regulatory environment is fragmenting rather than consolidating. With federal AI legislation stalled, states have become the primary drivers of binding AI rules. Colorado passed and then repealed its comprehensive AI law within months. Texas enacted the Responsible AI Governance Act limiting most requirements to government use of AI. California now has a layered compliance environment where multiple AI laws may apply simultaneously (VerifyWise, May 2026). A fragmented state-by-state regulatory landscape does not protect citizens more effectively than a unified federal framework. It creates compliance complexity that well-resourced agencies navigate and under-resourced ones ignore.

How Government Leaders Should Assess Their Actual Accountability Position

Five questions separate the government agencies deploying AI responsibly from the ones that will find out they were not at the worst possible moment.

1. Does the agency have a current, complete inventory of every AI system in use across all functions, including tools embedded in vendor platforms and commercial products procured through existing contracts?

2. For every AI use case classified as high-impact, does the agency have a documented risk management process that has been reviewed within the last 12 months and updated to reflect the current version of the system being used?

3. Has the agency assessed the outputs of its AI systems for disparate impact across the populations it serves, specifically whether automated decisions are producing different outcomes for different demographic groups?

4. Is there a formal human review process for AI-assisted decisions that affect individual citizens' access to benefits, services, or legal standing, and is that process documented and auditable?

5. If a citizen asked today how AI influenced a decision that affected them, could the agency answer that question honestly, completely, and in plain language?

An agency that cannot answer most of these is not running AI it fully understands. It is running AI it deployed, which in a government context means it is making decisions affecting millions of people through systems whose accountability structures have not kept pace with their deployment.

Bottom Line for Government Leaders

AI in government is producing real operational value. Agencies are processing higher volumes of requests with constrained headcount. Benefits are being delivered faster. Fraud is being identified earlier. Those outcomes are worth building on. But the speed of adoption has created an accountability gap that is already visible in the transparency reporting, the bias incidents, and the fragmented regulatory environment trying to catch up. By the end of 2026, government AI will be judged less by what is possible and more by what is dependable, reliable, auditable, and scalable (Granicus, April 2026). The agencies that build the governance structures to meet that standard now are the ones whose AI investments will compound into better public services over time. The ones that do not will continue expanding deployment while the accountability gap widens behind them. Cost is what agencies pay to deploy AI. Value is what responsible, equitable, continuously monitored deployment protects across every citizen decision, every public service, and every accountability conversation that follows. For government leaders, the ratio is not close. 

Works Cited

"Assessing the State of AI Adoption Across the Federal Government." Brookings Institution, 15 Apr. 2026, www.brookings.edu/articles/assessing-the-state-of-ai-adoption-across-the-federal-government.

"From Governance to Execution in Federal AI Policy." Brookings Institution, June 2026, www.brookings.edu/articles/from-governance-to-execution-in-federal-ai-policy.

"Artificial Intelligence: Generative AI Use and Management at Federal Agencies." U.S. Government Accountability Office, 29 Jul. 2025, www.gao.gov/products/gao-25-107653.

"How AI Is Quietly Reshaping Government Operations in 2026." Granicus, 6 Apr. 2026, granicus.com/blog/from-policy-to-practice-how-ai-is-quietly-reshaping-government-operations-in-2026.

"5 Ways State and Local Governments Will Operationalize AI in 2026." Route Fifty, 6 Apr. 2026, www.route-fifty.com/artificial-intelligence/2026/04/5-ways-state-and-local-governments-will-operationalize-ai-2026/412638.

"One-Year Retrospective on the Federal Government's Implementation of Updated AI Guidance." Center for Democracy and Technology, 22 Apr. 2026, cdt.org/insights/one-year-retrospective-on-the-federal-governments-implementation-of-updated-ai-guidance-accelerating-usage-with-incomplete-safeguards.

"Responsible AI Use in Local and State Government." American Association for the Advancement of Science, 2025, www.aaas.org/programs/epi-center/AI.

"U.S. AI Regulations 2026: The State Laws You Must Comply With." VerifyWise, 15 May 2026, verifywise.ai/blog/state-of-ai-governance-regulations-united-states-2026.