The Pentagon is launching a pilot program that will use artificial intelligence to manage how it classifies and handles sensitive military information, according to a memorandum issued by Deputy Defense Secretary Steve Feinberg that was reviewed by DefenseScoop. The directive orders an initial small-scale deployment of an automated security classification capability within the next six months — the opening move in an effort to overhaul a process the department describes as dangerously outdated.

Inside the Feinberg Memo For more context on this story, see our ongoing more AI stories.

The memo, issued Monday, frames the Pentagon's existing classification and declassification procedures as a source of "dysfunction" that the department says is "endangering" its mission. It argues that outdated procedures slow the "timely release of critical information" and erode public trust in the department.

"The dysfunction caused by these procedures leads to incorrect classifications that burden our national security systems, impede interoperability, and slows innovation by trapping vital data in stovepipes," the memo states, according to DefenseScoop.

ACME: The Air Force System at the Center

For the trial, the Department of Defense will use an AI-backed system the Air Force recently developed called the Automated Classification Management Environment, or ACME. The memo designates the Air Force's top civilian official as executive agent for the pilot, and an Air Force spokesperson told DefenseScoop that the service "will comply with the direction in the memo."

ACME was flagged publicly in August, when officials issued a request for information asking industry to help build a classification-management ecosystem to replace the "manual, human-intensive" process the military has used for decades. That notice described a system under strain: hundreds of officials hold authority to classify information, many using archaic programs, producing outdated documentation, high error rates, frequent over-classification, and degraded data integrity. The department also carries a hardcopy backlog of roughly 140 million pages.

From Pilot to Single Authoritative Reference

If the pilot succeeds, the memo says the automated system would eventually serve as "the single, digital authoritative reference" for all of the department's original classification decisions. That responsibility has historically belonged to specially designated human officials who determine how sensitive information is categorized — meaning a successful rollout would mark one of the most consequential transfers of a core government function to AI systems to date.

Why Officials Say the Change Is Needed

The case for automation rests on the sheer volume and velocity of modern military information. The August request for information called the manual process "incompatible" with the volume and speed of information in the digital age. Over-classification keeps information locked away from units and allies that need it; under-classification risks exposing capabilities and plans to adversaries. Errors in either direction carry costs, and with a seven-figure backlog of hardcopy pages, the department has concluded that incremental fixes are not enough.

Josh Wallin, a fellow at the Center for a New American Security's defense program, told DefenseScoop that under-classified information can expose a capability or plan to an adversary, while over-classified material can be withheld or stalled by a delayed review process — meaning it may not reach the organizations that need it, "in a coalition environment where information sharing is critical to coordination." He added that, in many cases, "the vast majority of information isn't necessarily so sensitive that you're worried about a misclassification," which is precisely why faster, more consistent handling could pay dividends across the department.

Analysts See Gains — and Real Risks

Wallin said automation could help with misclassification issues — including information that is over-protected, under-protected, or mislabeled by humans — and reduce stovepiping, especially as modern conflicts have shown that rapid information flow can be a decisive advantage.

But Wallin also flagged risks. Misclassification could occur at a faster scale under AI-backed systems. Faster sharing with allies and partners raises questions about the cybersecurity protocols of recipient networks. And accelerated declassification creates an "aggregate" effect, in which an adversary can piece together sensitive insights from individually harmless disclosures.

"There is a misclassification concern," Wallin said. "There is just an inherent concern that some information will be sensitive. Will it be misclassified and shared, sort of unintentionally, and will someone not catch it?"

Oversight Questions Remain

Wallin also questioned the brevity of the pilot and how human oversight will work when AI serves as the definitive reference for original classification authority. "How [are you] going to actually have human oversight in place? That's going to have to persist forever," he told DefenseScoop, noting that the system will evolve over time and that complete trust is never achievable.

The Pentagon did not immediately answer questions about which underlying AI models would power the pilot or whether the department has concerns about using the technology for this purpose. The directive lands amid a broader debate in Washington over AI oversight, even as defense officials argue that AI is too powerful a tool to sideline while adversaries deploy it across their own militaries. How the department balances those two impulses — aggressive adoption and meaningful human control — will be the defining question of the six-month pilot.

For the Air Force, the assignment is also a vote of confidence. The service built ACME while the department-wide conversation about classification reform was still theoretical, and the memo now hands it responsibility for a capability that, if it works, will reshape how every military branch handles its secrets. For the rest of the federal government — where classification backlogs and over-classification complaints are chronic — a successful Pentagon pilot would provide the first large-scale template for AI-assisted information handling, and a closely watched one at that.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →