Peter Hamlin/AP Illustration
When the “Department of Government Efficiency,” proposed by shadow president Elon Musk, was reimagined from an outside advisory group making recommendations on federal spending to a temporary organization inside the U.S. Digital Service, a division of the Executive Office of the President, I thought it was partially a way to avoid transparency and disclosure rules for advisory commissions. The U.S. Digital Service’s mission is to modernize and improve government technology and software, and a lot of people scoffed at the idea that a plan to cut trillions of dollars in spending would suddenly devolve into an IT upgrade.
But if you read the executive order that made this transformation, it outlines DOGE’s task clearly. There’s a little clause in there ordering agencies to grant DOGE “full and prompt access to all unclassified agency records, software systems, and IT systems.”
That’s pretty much what DOGE has been doing, storming databases and systems across agencies. We’ve talked a lot about the control of the Bureau of the Fiscal Service, the government’s primary payment system. They’ve gained access to records at the Small Business Administration, and federal employee data at the Office of Personnel Management, and scientific data at the National Oceanic and Atmospheric Administration, and student data at the Department of Education, and the IT systems of the Labor Department and General Services Administration. ProPublica reports that DOGE will look at the Central Accounting Reporting System, a database that tracks financial flows across the government. If there’s a federal agency or piece of software not on that list, they’re probably coming for it soon.
Much of this is either possibly or definitely illegal, and so potentially damaging to governmental functioning that one agency has labeled it an insider threat. Lawsuits have been filed, and injunctions and limitations have been issued in some cases. But the question that hasn’t been totally answered is: Why does Musk want this access?
We have some answers to this multifaceted question. Four days after the inauguration, Tom Krause, the private equity hatchet man Musk brought in for his Treasury project, demanded access to the Bureau of the Fiscal Service’s payment system so it could pause all disbursements to the U.S. Agency for International Development (USAID). He was rebuffed by the acting Treasury secretary, longtime nonpartisan employee David Lebryk, who said there is no legal authority for Treasury to halt an entire agency’s payments. When Scott Bessent was confirmed as Treasury secretary, Lebryk was forced out, DOGE was granted payment system access, and Krause now has Lebryk’s job.
When you automate government tasks, they’re only going to be as good as the engineers doing the automating and their knowledge of the federal government.
So gaining control of the nation’s fiscal plumbing was certainly part of the plan, as part of the larger bid to centralize budgetary control in the hands of the president, and effectively seize control of Congress’s power of the purse. This is alarming and dangerous. But DOGE is also looking at personnel files, scientific information, agency databases, and more. What about that?
There are a couple of working theories. One is old-fashioned opposition research. Musk needs a steady stream of propaganda and slander to post to X to whip up feeding frenzies about employees or government programs. This data could launch a million tweets.
The early examples we’ve seen of this show that the data takeover was both unnecessary and a precursor to rank stupidity. Musk slandered a Department of Agriculture official as a DEI hire when she was working on crop diversification, not racial diversity. A Musk fanboy querying usaspending.gov found government “funding” of Politico, which sounds odd until you realize it’s just federal employees buying subscriptions to Politico Pro, the paywalled insider-news service that produces a lot of high-quality reporting.
Both of those examples were pulled from the transparent federal databases that are already public, calling into question the need to take further control. Meanwhile, the Politico thing has already led to a White House cancellation of reimbursements for the subscriptions, which is fine (the executive branch can determine its own reimbursable expenses) if you think the $35 trillion national debt will be “solved” with $8.2 million in subscription rollbacks, which represents the tiniest of fractions of the federal budget.
In other words, part of this alleged reinvention of government we’re seeing involves the same thing conservatives have been doing for a half-century: dig up dirt on microscopic funny-sounding programs while ignoring the real drivers of wasteful federal spending, like the trillions of dollars that get funneled up from ordinary taxpayers to elites like Elon Musk.
But there’s another reason DOGE is after the IT systems. They want to integrate artificial intelligence into every facet of government operations, both to automate aspects of government work and to flow data through AI to catch discrepancies. Thomas Shedd, a former Tesla employee now serving as a top official at the General Services Administration (GSA), called it an “AI-first strategy” in an all-staff meeting. GSA is developing a custom chatbot to analyze government procurement and contract data, and DOGE teams at the Department of Education have also pushed that department’s data through AI software.
So you have a bunch of techno-futurists who think that the future of government efficiency is either cherry-picking stuff that sounds bad, or running everything through a supposed HAL 9000 super-computer to weed out the waste. Neither will accomplish their aims, and will probably lead to tons of negative consequences in the process.
First of all, when you automate government tasks, they’re only going to be as good as the engineers doing the automating and their knowledge of the federal government. Already, we’ve seen evidence of outrageous incompetence. DOGE’s “deferred resignation” offer, which was made to most federal employees, also went to federal judges, who have lifetime appointments. (The offer was put on hold temporarily in the courts last week; not sure if that judge also got the offer.) In another case, every employee hired at the CIA in the last two years was put into an unclassified document, almost surely using some automated tool.
A smaller point is that using AI services across the government will be a financial bonanza for the companies chosen, and clear and obvious conflicts of interest abound. Musk has an AI development firm, for example.
But the biggest issue with using AI is twofold: the routine wild inaccuracy of AI tools, as well as the particularly haphazard manner in which DOGE wants to use it. The Biden administration had been slowly integrating AI into government tasks, but it was pretty clear about guarding against potential risks: errors in coding, for example, or added security vulnerabilities. The Musk team is not going to care about that at all. If a one-line code error flags billions of dollars in “wasteful” spending improperly, that’s probably seen as a feature, not a bug. If the AI goes haywire and starts making things up, or the code breaks systems that need to be redundant and reliant, oh well, government isn’t supposed to work well anyway. And if back doors are opened up to allow hackers or adversaries to poke through, maybe they’ll develop a patch later.
That’s the Silicon Valley mindset: move fast and break things, like Social Security. And it has merged with the MAGA mindset. Without transparency in what AI is supposed to look for, you can bet it will be reverse-engineered to spit out a desired result, whether about climate change or DEI or whatever.
So on one level, you can expect the inexperienced coders trying to find trillions in cuts by lazily feeding government data through AI tools to make mistakes, trigger cybersecurity breaches, and neglect AI hallucinations. On another level, you can expect government data to be highly politicized. The White House plans to make chief information officers at the agencies political appointees, who would have loyalty to the president and a willingness to manipulate data for ideological purposes. We’ve already seen this with the installation of Musk ally Tom Krause overseeing the Treasury payment system.
As data is hoarded inside the government, it’s disappeared to the outside world. Websites across the government were taken down, de-wokified, and sometimes erratically restored. This is data that is used to make decisions in the economy and public health and much more. And if the public doesn’t have it, widening that gap between who knows what’s going on and who doesn’t, the results can be catastrophic.
In fact, there are many real-world implications to the great software putsch. The system could really break and your grandmother won’t get her Social Security check on time. But even without that apocalypse, we may have a government run by a malicious autopilot, blind to its own errors and bias. It absolves Musk, and more so Donald Trump, from blame for the conclusions drawn: The AI said that poor people need less food assistance, see, not any human being. It seeks to put government in the hands of whoever controls an algorithm rather than an elected representative of the people. It couldn’t be more at odds with democracy.
