How the Defense Intelligence Agency is fast-tracking data and AI modernization

Defense and intelligence agencies play a crucial role in supporting the military by detecting threats before they cause real harm.

The approach isn’t confined just to the military. National security, cybersecurity, and law enforcement organizations similarly focus resources “to the left of boom” — leveraging intelligence gathering, surveillance, and analysis of adversaries’ behavior to plan for potential disruptions or catastrophic events.  

For agencies like the Defense Intelligence Agency (DIA), however, those efforts have become increasingly challenging. One reason is the pressure to manage and process ever-increasing volumes of global intelligence data. Another is the need to keep up with and adapt to the unprecedented pace and scale of technological change, said DIA Chief Information Officer E.P. Mathew, at a recent GDIT Battlespace of the Future summit, hosted by Scoop News Group.

During a panel discussion with Aaron Bedrowski, Senior Vice President at GDIT’s Intelligence and Homeland Security Division, Mathew highlighted the accelerating pace of technology adoption, noting that ChatGPT reached 50 million users in just five days. YouTube reached the same milestone in 10 months. By comparison, radio and television took decades. In a similar vein, the compression of Moore’s Law — the time it traditionally takes to double semiconductor capacity while halving computer costs — has fallen from roughly every two years to seven months. 

Central to DIA’s challenges is the extent to which the government’s traditional five-year budgeting and planning cycles and acquisition timelines are failing to keep pace with enterprise hardware and software upgrades. Mathew noted that over five years, the capabilities of a semiconductor chip can improve by as much as 1,000 times. That’s making it harder for agencies to plan, acquire, and deploy up-to-date technologies, and it adds to the growing risk of national security vulnerabilities, he explained. 

Modular component platforms key to AI

To survive technology’s accelerating innovation cycle, Mathew maintained that defense leaders must undertake a comprehensive cultural overhaul, shifting away from infrastructure-heavy, application-centric frameworks toward agile, secure, data-centric environments capable of continuously adopting commercial software upgrades.  

“Our goal is to build a data environment where the focus is on you having access to data by policy only,” Mathew explained. Rooted in the Pentagon’s zero-trust cybersecurity framework, access to information is restricted by default. Every ingested dataset is meticulously tagged, cataloged, and encrypted. Users are verified and granted access to data dynamically based on centrally managed policies. Crucially, this security envelope remains intact even if the data leaves its core repository, with periodic retesting of user credentials to ensure absolute privacy and operational integrity.

This structured, policy-driven data layer also serves as the mandatory technical foundation for AI and advanced automated computing, he said, adding that an organization “cannot do AI unless you do data correctly.”

Another challenge agencies face, noted GDIT’s Bedrowski, is how to integrate and scale emerging technologies faster. New capabilities happen “so quickly that by the time you’re in the integration, something new comes out,” he noted.

To scale AI and emerging technologies into existing intelligence operations and enhance decision-making, Mathew said the DIA follows a three-step process:

• First, data must be structured and restricted through granular, centrally managed policy entitlements.
• Second, the agency is implementing a Modular Component Platform (MCP) to allow diverse components to pull data simultaneously.
• Finally, DIA’s network must integrate semantic AI capabilities — including knowledge graphs, entity (data-level) resolution, and graph (structural) resolution — to turn raw, disparate streams of information into meaningful operational intelligence.

By engineering systems using this highly modular blueprint, DIA expects to gain structural agility to quickly swap out individual software components as superior commercial tools emerge. This approach prevents defense networks from being locked into long-term proprietary vendor contracts. It also enables a more seamless evolution from basic decision support to more comprehensive decision augmentation and automation.

Overcoming workforce deficits through immersions

Ultimately, configuring and deploying these highly modular architectures requires specialized technical leads who understand how to tailor emerging technology to complex national security missions. Mathew acknowledged that building internal expertise has been severely hampered by a recent Defense Reform Program (DRP) reallocation that led to the departure of roughly 22% of DIA’s specialized network and software engineering workforce.

To counter this talent drain, the DIA has begun bypassing traditional post-sales vendor support models, which Mathew characterized as deeply inadequate for long-term operational success. Rather than purchasing commercial software and relying on external contractors for maintenance, the agency has created a hands-on internal training lab and launched an innovative “Training with Industry” initiative.

Borrowing a page from a similar Army initiative, this immersion program embeds DIA personnel directly within commercial technology companies for six-month rotations. By gaining direct experience with leading-edge software platforms, these operators return to the agency better equipped to serve as internal product leads who can maximize existing software investments, Mathew insisted.

Additionally, the agency began reaching out to industry earlier this year to secure new capabilities to test, evaluate, verify and validate new AI technologies.

This article was produced by Scoop News Group for DefenseScoop and sponsored by GDIT.