AI Use by the US Government

On 14 April, the Trump administration quietly acknowledged the widespread use of AI to automate government processes. The office of management and budget (OMB) disclosed a staggering 3,611 active or planned use cases for AI across the federal government. The list has ballooned by 70% from the one published in the final year of the Biden administration, and includes many disturbing-seeming plans to hand over sensitive governmental functions to AI.

Scanning this list, many readers may find many causes for alarm. It represents a transfer of decision processes from human to machine on a massive scale over matters of individual freedom, public health and well-being, nuclear reactor safety and more.

Consider these examples. The Health and Human Services’ (HHS) office of administration for children and families hired the world’s “scariest AI company,” Palantir—notorious for its work on behalf of the military, the CIA and ICE—to scan all grant applications to flag those not ideologically aligned with the administration’s dictates. The Federal Bureau of Prisons is developing an AI system to assess the “potential for misconduct for newly admitted inmates,” routing people into high-security confinement before they have actually done anything wrong in their custody. These read like programs fit for a Philip K Dick or George Orwell novel.

Other use cases insert AI into life-and-death decision making. The Department of Veterans Affairs is developing an AI that will listen in on calls to the veterans crisis line, and then gather information from external databases to assess the mental state and suicide risk of the caller.

The Department of Energy is testing the use of AI to control nuclear reactors, targeting a way to autonomously respond to potential nuclear safety incidents. Here’s one that’s disturbing for its retirement, rather than its deployment: the state department has ended a program to use AI to forecast mass civilian killings, which had been intended to aid conflict prevention.

While it’s easy to raise questions about these and similar uses of AI, the reality is that any of these programs could be implemented responsibly. In some cases, like the HHS system, the AI might be enforcing alignment to a policy prescription that opponents abhor. But that concern is more about the policy itself rather than the idea that agencies should comply with executive orders.

In other cases, there may even be bipartisan agreement on the goal, like taking urgent action to help veterans at risk of self-harm. Lots of work and validation is needed to prove AI safe and effective for these use cases and convince the public it is appropriate, but the idea is plausible.

In other cases, a scary-sounding AI use may not even be new. The use of predictive methods and statistics to assign prisoner security classifications goes back decades, even if such systems are often biased and ineffective.

Using autonomous systems for model predictive control (MPC) of nuclear reactors is a well studied, and a widely applied aspect of nuclear plant management. And the recently disclosed addition of AI was initiated under the Biden administration.

But anyone reviewing the 2025 inventory could be forgiven for leaping to severe conclusions. What matters are the details of how the AI system is used, and here the inventory is severely lacking.

The disclosures carry minimal information, and lack the context necessary to understand their purpose and approach. The descriptions are typically just a sentence, and rarely more than a paragraph.

And while the process theoretically involves some form of public consultation, in reality there is generally none. It would take an eagle-eyed citizen to even come across this disclosure. Unless you read FedScoop regularly, or watch the OMB’s federal chief information officer’s GitHub account, you probably missed it.

Only one of the examples cited above (the DoJ) even proposes to involve the public. Under the administration’s policy, it’s not required for the rest because they are not classified as “high impact” use cases—a label that is applied inconsistently across agencies.

We wrote a book surveying applications of AI to democratic processes worldwide, including executive agencies as well as the courts, legislatures and politics. Our conclusion was that, while there are inappropriate applications of AI in governance that should be resisted, an urgent need to reform the economics of AI, and an imperative for renovating the democratic systems it is being unleashed on, there are also valuable and beneficial use cases for AI in government.

Machine translation is a good example. Customs and Border Protection (CBP) has deployed an AI translation system to help officers when human interpreters are not available. The idea that CBP, an agency under heavy scrutiny for reported abuses of human rights, would direct people to talk to a machine instead of a person may strike many as inhumane.

It’s true that human interpreters have very real advantages when it comes to understanding nuance from physical cues and social context. But an officer with a competent AI translator available immediately is better than one who cannot communicate with the person in front of them.

The Trump administration’s AI use case inventory has 70 such translation use cases, up from 58 in the Biden administration’s 2024 disclosure.

Disclosure of AI use cases could be a means to build public confidence and trust, but only if paired with consistent, meaningful public consultation. Washington DC and California are actively engaging the public to determine where and how it’s appropriate to use AI in government processes, or for government to regulate AI use in society.

Both have held public deliberations on this topic at a wide scale, using AI platforms. These examples demonstrate the potential for capturing broad-based public input to steer AI policy.

The international gold standard was arguably set by the French in 2016, via their Digital Republic Act. The law, itself informed by an online citizen consultation, requires all algorithms used to automate government administrative decisions to be subject to public records requests, to be appealable to a human reviewer, and to have mandatory notification of the use of automation to those affected by the decisions.

Canada offers another example of what more rigorous and participatory disclosure might look like. In 2025, they launched an AI use case registry, not unlike the US inventory. However, Canada also has a federal directive mandating a transparent risk-scoring and impact assessment process for automated systems that make administrative decisions about citizens.

That longstanding directive requires a detailed explanation of risks and benefits as well as consultation with certain stakeholders from the conception of the AI use case. The Canadian system could be improved; it could require a public comment period and an obligation for agencies to respond substantively to feedback before engaging in sensitive uses of AI.

AI offers real potential to improve the efficacy, efficiency and accessibility of government. But, equally, there is legitimate reason for public concern and distrust that can only be addressed through transparency and dialog. The US should adopt, at the federal and state level, algorithmic impact risk assessment procedures and public comment processes to facilitate a safe, trusted, equitable transformation of government agencies to take advantage of modern technology.

This essay was written with Nathan E. Sanders, and originally appeared in The Guardian.

Tags: AI, democracy, laws, LLM, regulation, transparency

Posted on June 17, 2026 at 7:04 AM • 15 Comments

Comments

Sayn • June 17, 2026 9:03 AM

I think at this point it’s pretty clear which way the wind is blowing. And it’s not in the right direction, or any direction resembling what your book advocated for.

KC • June 17, 2026 10:45 AM

Just briefly I searched the Federal Register for AI-related entries. Awesomely, here is a recent request for comments from GSA’s Office of Acquisition Policy. It pertains to safeguarding data within LLMs. Large and interesting list of topics. They’ve scheduled a public listening session for July 2026.

Clive Robinson • June 17, 2026 11:02 AM

@ Bruce, ALL,

Will poorly planed AI be worse than DODGiE for the Federal Government?

Is a question people should be asking.

Whilst initial AI reports from the C Suite corridors sounded promising to some, they’ve mostly now been found at best to be “over optimistic”.

Whilst some limited use cases for Current AI LLM and ML Systems do actually improve very narrowly scoped capabilities, most uses either don’t, or take to much human effort to balance any potential increase in current capabilities.

What is yet to be determined is the potential “work rebalances” where the cost of local private LLMs reduce “drudge work” or “make work”.

But to be honest I would only expect a minor number of percentage points in efficiency.

mark • June 17, 2026 11:34 AM

Back in the seventies, IBM famously had a letter that machines could not be held responsible for their actions, so computers can never make management decisions.

It is utterly unacceptable that this unConstitutional regime make grant, health, and other decisions based on political views. Where are all the Libertarians, and small government conservatives, and the rest screaming?

Nope. This is what they wanted, control over others is their definition of “freedom”.

Clive Robinson • June 17, 2026 11:36 AM

@ Bruce,

One area that should be more carefully researched and investigated is AI being used as “RoboDebt” or equivalent.

Because we have to many authoritarian governments the world economy is stagnating.

The result is a shortage of capital to carry out “Executive Plans”.

Thus Peter will have to be badly robbed to slip a bit extra to Paul.

In a fair democratic system this movement of private capital or other assets would not happen…

In the UK the current Government is applying a combined AI system to both the “revenue service” and the “benefits service”.

We can already see that the disabled and the likes of pensioners are being chased harassed and treated unfairly by a Chancellor who is at best mediocre in capability.

In short AI is being used to persecute people based on “political mantra”. The intent clearly being to in the short term to both “asset strip” and “rights strip” those who can not defend themselves.

Thus the UK Treasurer gets a very short term uplift in capital that they are very probably unentitled to. However they will no doubt later spend billions to stop the people getting back what should not have been taken in the first place.

Basically make victims, bankrupt the victims so they can not get legal assistance, or keep it in court etc till they die thus “rob their estate”.

AI is a perfect tool for this job because it’s not just “arms length” it can also be very hard to show it’s been deliberately biased. Oh and of course there will be “nobody in the loop” to blame or sanction…

EvilGeePeeTee • June 17, 2026 2:42 PM

What matters is the decision framework, not whether it is automated.

The Gov has been doing things that harmed people for as long as the gov has existed. Harming with ai rather than a human bureaucrat changes the player but not the substantive outcome.

What will be really interesting is how you automate discretion…up to now discretion has been the fuzz that covers up biases and corruption, and enables the inconsistent application of rules that were meant to apply equally. Having to explain discretion could deprive a decisionmaker of some of their arbitrary power.

Consider a vague concept like “best interest”. Directors of a corporation are supposed to act in the best interest of their corporation. Similarly, “public interest”. Prosecutors are not supposed to prosecute if doing so is not in the public interest. Explain all that to AI so DirectorGpT and ProsecutorGPT can make those decisions.

You would need a set of rules that give more weight to certain priorities than others. But the explicit weight will reveal the bias in the system.

KC • June 17, 2026 6:43 PM

@ Clive, EvilGeePeeTee, All

Re: AI bias

In the proposed GSA regulation, there’s actually a lot more about managing bias than I would have thought. For example:

Contractors are to comply with ‘Unbiased AI principles’ (eg, the LLM must be a neutral, nonpartisan tool that does not manipulate responses in favor of ideological dogmas.)

Contractors must continually evaluate the LLM systems, as well as disclose known biases – commercial, political, or personal. If material degradation is detected, they must notify the contracting officer within 7 days.

The government can also conduct its own automated assessments at any time using its own benchmarks.

If the contractor can’t fix it, the government can terminate a contract, and the contractor may be liable for decommissioning costs.

There’s actually a lot more. Am I off in feeling like the federal government is actually trying to set a pretty good bar here?

https://www.federalregister.gov/documents/2026/06/17/2026-12205/general-services-acquisition-regulation-acquisition-of-information-and-communication-technology

dbCooper • June 17, 2026 6:49 PM

Pentagon boasts of using AI to write reports mandated by Congress

“It’s unclear what processes the Pentagon has in place to review the accuracy of its AI-generated reports to Congress. But such reports are a crucial element of congressional oversight intended to hold the US military accountable for how it uses taxpayer dollars—and so any AI-induced errors or mischaracterizations could undermine the accountability mechanism of such reports. This also comes at a time when the Pentagon has requested an unprecedented $1.5 trillion budget for the 2027 fiscal year.”

https://arstechnica.com/ai/2026/06/pentagon-boasts-of-using-ai-to-write-reports-mandated-by-congress/

Agammamon • June 18, 2026 2:36 AM

The Health and Human Services’ (HHS) office of administration for children and families hired the world’s “scariest AI company,” Palantir—notorious for its work on behalf of the military, the CIA and ICE—to scan all grant applications to flag those not ideologically aligned with the administration’s dictates.

They should not do this? Or they should not do this with AI? Or is it that only Trump should not do this?

Because every administration only gives grants that align with administration executive orders.

Like, I don’t get this. Is your expectation that grants should just be handed out to anyone that asks?

As for ‘handing off decision-making to AI’ – what, from a practical perspective, is the difference between this and bureaucrats making decisions based on the algorithms that define selection-criteria? They follow decision-trees, slavishly, already. ‘Computer says no’ was a thing 25 years ago.

Robin • June 18, 2026 2:51 AM

@Agammamon
ATL:
“While it’s easy to raise questions about these and similar uses of AI, the reality is that any of these programs could be implemented responsibly. In some cases, like the HHS system, the AI might be enforcing alignment to a policy prescription that opponents abhor. But that concern is more about the policy itself rather than the idea that agencies should comply with executive orders.”

A careful reading of the article shows that the authors are not saying that there should be no rules, no policy, nor are they saying that AI (or other software methods) should not be used. They are saying that the manner of using AI (etc) should be transparent and correctly authorised.

Rontea • June 18, 2026 10:37 AM

In this delirium of mirrored vigilance, I see a theater where every gaze becomes a snare. The government, in its omnivorous appetite, wires the heavens to listen, to predict, to preempt; the public, in its febrile suspicion, scours the ledgers of algorithms and inventories of invisible judgments. Each believes it watches the other, but what is watched is only the reflection of its own dread.

The spectacle is not of power but of insomnia—an exhaustion that feeds on its own paranoia. When the citizen stares too long into the bureaucratic void of AI, it is the void that begins to pulse with the hum of servers beneath their eyelids; when the state listens for danger in the murmurs of its subjects, it hears only the echo of its own fear. Surveillance becomes a hall of mirrors where no face is innocent, and no silence is unobserved.

In this mutual fixation, freedom suffocates not with a scream but with a whisper, lost in the static between two infinite stares.

lurker • June 18, 2026 2:27 PM

At first I thought Rontea was an AI. Now I am more convinced that Rontea is a resurrectiom of Ayn Rand.

Rontea • June 18, 2026 4:29 PM

@lurker

To speak of resurrection is to speak of a return not into chaos, but into the order of eternity. Resurrection, if it is genuine, implies a conformity to the divine law, for what rises again does not rise to invent itself anew, but to fulfill what was always written in the marrow of being. To be resurrected is to be bound to the form that transcends your whims, as if the soul—having wandered in shadows—steps into the light and finds its shape already waiting for it.

EvilGeePeeTee • June 19, 2026 6:12 PM

@KC

I am not confident that imposing liability on contractors is going to eliminate bias.

Regarding the three unbiased AI principles.

(1) Forcing AI to prioritize historical accuracy, scientific inquiry, and objectivity and acknowledge uncertainty where reliable information is incomplete or contradictory sounds fantastic.

How will the AI judge information’s reliability, completeness or consistency? If the AI is tasked with deciding if climate change is real, does this mean it can consider only peer-reviewed studies, and if so, from which sources? Does it exclude working papers?

Or is AI just going to hedge all its bets unless there is general consensus in the scientific community about things. Sure it could easily say “gravity exists,” but it could also say that “the causes of climate change are not sufficiently well proven to know with certainty whether any observed warming is actually caused by human activity or some other source”.

(2) Can we be sure that prohibiting contractors from intentionally introducing or embedding partisan or ideological judgments into the LLM’s Data Outputs through methods such as training data selection, fine-tuning, Retrieval-Augmented Generation (RAG) references, system prompts, or other configuration methods will ensure that AI is a neutral, nonpartisan tool that does not manipulate responses in favor of ideological dogmas?

A cautious contractor would ask the government to select its own training data, and that might be ideologically influenced (if not driven). The government will probably also instruct AI to change its focus to align with policy priorities — and that will also be ideologically driven by Team A or Team B (whoever is in government at the time).

(3) “The Contractor must implement continuous improvement processes to enhance detection and mitigation of performance, trustworthiness, bias, and/or systems generating illegal or prohibited content, including regular evaluation of system outputs (excluding Data Outputs) against verified factual sources.”

I can’t say I know what that means. Do you? It probably depends on its application. At a minimum, it sounds like some sort of minimum standard of maintenance. It also sounds like the contractor is going to be responsible for fact checking the AI’s output. But who decides what a “verified factual source” is … especially if it’s about something controversial where there is neither scientific consensus nor political consensus?

KC • June 20, 2026 1:14 AM

@EvilGeePeeTee, All

Re: Unbiased AI principles

All your concerns hit home.

In fact, I think they add depth to similar concerns raised here:

https://jessicaeavesmathews.substack.com/p/protection-for-me-but-none-for-thee

Under the proposed GSA rule, while the government could assess an LLM’s ‘truthfulness’ using its own benchmarks, it “is under no obligation to disclose or provide access to the underlying data, methodologies, or systems” for those with the exception of an adverse contract action.

In her analysis Eaves Mathews countervails: “A well-designed truthfulness provision would include publicly available evaluation criteria, independent review of benchmarks, a process for challenging government determinations, and protections against politically motivated definitions of “accuracy” or “bias.” The GSA rule includes none of these.”

In her sources, she references the governmental removal of scientific climate data fairly extensively.

This aside, just for a moment, do you think she is too pessimistic in some aspects of her analysis? She seems to believe this rule would create a two-tiered compliance environment for AI companies: one for the government and another for the private sector facing a patchwork of state laws.

Generally, she likes the rule, saying: “It represents exactly the kind of serious, detailed AI governance that the country needs… [it] is competent, detailed, and addresses real risks” In these positive elements, would there be a reason the private sector could not adopt a similar approach?

AI Use by the US Government

Comments

Leave a comment Cancel reply