Pentagon Chief Technology Officer Emil Michael said in a recent event with Hudson Institute, a conservative think tank, that the use of AI across the agency has proliferated, revealing that 80,000 personnel were using commercial AI tools as of December 2025, and that number has grown to 1.5 million personnel as of this month. The Department of Defense has about 3.5 million employees, meaning less than half, roughly 43%, of agency employees are using AI.

It’s a loaded topic, not least because the short-lived stint of Elon Musk at the Department of Government Efficiency (DOGE) started with grand claims of AI efficiencies that led to mass layoffs and not much evidence of anything else. A closer look into Michael’s claims suggest that the adoption is surging, yet struggling in ways that mirror wider adoption problems inside the Fortune 500.

Government AI use has definitely ballooned in recent years, with the Office of Management and Budget disclosing more than 3,600 active or planned use cases for the technology last year, a 70% increase from the year before. Beyond generating Congressional reports, these use cases include The Federal Bureau of Prisons developing a system to assess “potential for misconduct for newly admitted inmates” and assign security levels to incarcerated individuals. Court documents reveal DOGE used ChatGPT to flag diversity, equity, and inclusion initiatives receiving federal funding and cancelled one museum’s $350,000 grant to replace its heating, ventilation, and air conditioning (HVAC) system.

AI deployment in the agency snowballed following the creation of DOGE, which effectively replaced the U.S. Digital Services, an Obama administration-era initiative to modernize agency technology. In 2025, Google and other U.S. tech companies signed agreements with the U.S. General Services Administration (GSA) to make their tools available to federal agencies at discounted rates. In May, the Pentagon announced a series of partnerships with SpaceX, OpenAI, Google, NVIDIA, Reflection, Microsoft, Amazon Web Services, and Oracle for operational use of their respective AI tools.

For decades, the Pentagon has led the rest of the federal government in integrating AI into its work. The government introduced AI initiatives to agencies as early as the 1960s, with early experiments of the technology used to solve logistical issues in the Department of Defense, but the AI in Government Act of 2020 during the first Trump administration jumpstarted the practical applications of the technology, with resources being distributed by the GSA. 

The Pentagon has other issues, as Fortune senior contributing columnist Steve Hanke has noted: it has not ever successfully passed an audit. Is AI use really enough to overcome that kind of internal dysfunction?

The Pentagon’s victory against bureaucracy

If one of DOGE’s goals was to leverage technology to decrease time spent on bureaucratic labor, Pentagon officials say they’ve notched a victory. Among the integration of AI tools is writing mandatory reports for Congress. According to Michael, the use of the technology has cut down the time to produce these reports exponentially. The number of mandated Department of Defense reports to Congress has swelled from 500 in 2000 to about 1,400 in 2020 as a result of new defense appropriations bills, per data from the U.S. Government Accountability Office (GAO).

“I have to report to Congress every year on this thing,” Michael said at the event. “Let me load all the papers onto it and have it draft me a congressional report that would otherwise take 200 hours of staffing time and do it in five hours.”

It’s not the first time Pentagon officials have praised AI for saving time on these reports. Earlier this year at a Box Federal Summit event, Jacob Glassman, the Pentagon deputy assistant secretary of defense for science and technology foundations, said that he asked a short-staffed team to use GenAI.mil, the Pentagon’s enterprise AI tool, to generate a report for Congress, DefenseScoop reported. Glassman did not specify the report the tool generated.

“Sure enough, they came back to me about a week later and said…‘Not only did we generate the report, [but] this is the best report we’ve written in the past five years.’” Glassman said. “So I’m like, this is incredible, right?”

The agency did not respond to Fortune’s request for comment. 

The risks and challenges of government AI use

Broader reports on AI adoption in the government shows a slower adoption story, full of challenges and stopgaps. The Brookings Institute analyzed federal jobs data, OMB memoranda, and interviews with current and former technologists across eight federal agencies through the end of 2025. Its conclusion was that over the previous three years, five large federal agencies accounted for more than half of the government’s total reported AI inventories use. The 11 small federal agencies reported just 2% of the total AI inventory use in 2025.

Though data to track AI adoption in the federal workforce has been inconsistent, Brookings explained that “workforce capacity constraints, a risk-averse culture, procurement and funding challenges, and low public trust in AI systems” have slowed adoption efforts. With few resources to train or familiarize staff with AI, agencies are often left with a workforce frustrated or unequipped to experiment and use it. 

It’s a similar story in the public sector. A recent Goldman Sachs’ March 2026 AI Adoption Tracker indicated that while AI adoption is growing, it’s not doing so at the rate deemed necessary for it to be a standard workplace tool. A whopping 80% of companies were not adopting AI across the economy, and the adoption rate actually ticked down to 19.5% as of May 2026, per the Census Bureau’s Business Trends and Outlook Survey. Adoption is only expected to hit 22.7% over the next six months. Establishments with over 250 employees continue to lead in adoption, and they are only at 36.1%.

Analysts noted that while AI is saving workers up to an hour per day, productivity increases are modest to none because adoption isn’t high enough. SAP subsidiary WalkMe’s annual State of Digital Adoption report in April found that more than half of workers surveyed bypassed AI tools to do the work manually instead. Only 9% of workers said they trust the technology for complex tasks.

There are risks associated with the plunge into AI as well. A GAO report from March revealed that among government and industry experts the agency spoke to, many believed that while AI could make it easier to cross-reference government data and make it more accessible across departments, it can also be used to intentionally or unintentionally general false information. Increased access to data can also mean agencies, businesses, and other organizations could access data beyond their intended purposes, such as a company using tax return information to determine product prices.

The watchdog called on OMB to identify privacy-related risks associated with increased AI use and provide guidance on considerations agencies should make when deploying these tools.

“Without providing this additional information,” the report warned, “agencies are at risk of potentially exposing sensitive information that can negatively impact the public.”