An AI-powered replacement for the repository of intelligence on foreign militaries is headed for the Pentagon’s classified network, where it will help pioneer new security techniques and become fully operational about a year later than planned, officials said.
“We expect” the Machine-assisted Analytic Rapid-repository System to be approved to run on the SIPRNet “within the next few weeks, actually,” said Doug Cossa, chief information officer of the Defense Intelligence Agency, speaking at an Intelligence and National Security Alliance virtual event on Tuesday.
MARS is to replace the Modernized Integrated Database, or MIDB, the Pentagon’s main repository for “foundational military intelligence”: the information collected by the intelligence community on other countries’ militaries and infrastructure. The DIA began working on MARS in 2018, adding new tools and AI-powered capabilities along the way, but it has only been accessible in certain locations. An agency spokesperson declined to say where.
The move to SIPRNet will make MARS far more widely available. Full operation is expected in 2026, the spokesman confirmed, about a year later than originally planned.
The intelligence community has been working to improve their ability to sense, aggregate, display, and share data. But with each agency and military branch working on their own efforts, integrating them to form a common operating picture is an ongoing challenge.
“The challenge we have ahead of us is: how do we actually integrate that data into one common operating picture? And this is a priority of not just the [director of national intelligence], but the [under secretary of defense for intelligence and security] of collaborating across the intelligence and defense enterprise to create that common operating picture to where we are all leveraging each other’s data in an authoritative way, but the same sheet of music in a synchronized fashion,” Cossa said.
The Pentagon asked for $890 million in 2024 and $977 million in 2025 to upgrade its networks to zero trust architecture. Those funds will be used to spur several pilots over the next year despite limited cyber personnel for vulnerability testing, said Randy Resnick, who leads the Pentagon’s zero trust office.
“We’re pretty complete with all the foundational documentation and how it glides or fits into the authorization to operate process, and now we’re doing pilots,” Resnick said Wednesday during a panel discussion at NVIDIA’s AI Summit in Washington, D.C. “I’m expecting some successes, many successes, which will show that we can hit what we’re calling ‘target’ or ‘advanced level’ zero trust. And that whole strategy is meant to stop the adversary.”
To meet the Pentagon’s 2027 deadline for implementing zero trust—the principle that focuses on securing data and that no device or user should be automatically trusted, Cossa said the agency put much of its zero trust “investments” for 2024 and 2024 in MARS.
Moreover, he said MARS will be the first DIA system to offer “fine-grained entitlements,” a way to customize and manage access down to individuals and devices.
These and other security measures could help the intelligence community and Defense Department inch towards a common operating picture when it comes to military intelligence.
The goal is to ensure that data pulled from a myriad sources across the IC and DOD is secure and can integrate with MARS. Moreover, employing zero trust as a baseline for defense and intelligence systems could be the key to frictionless information sharing with allies and partners.
“It means that we don’t need these disparate networks anymore. We can actually host data in one place and regulate and control and secure access through those fine-grained entitlements for both users and devices. That’s the future of where we’re going,” Cossa said. “When we say we’re going to put MARS on SIPR, yeah, we’re going to. But what we want to see in the future—and really the objective—is we don’t have to deploy it in different networks. We deploy in one area, and we manage the accesses through entitlements.”
