NGA Expands AI Use While Managing Unrealistic Intelligence Expectations

The National Geospatial-Intelligence Agency is accelerating artificial intelligence deployment across its operations, even as senior officials push back against growing expectations for constant, real-time intelligence delivery. The agency operates within the Department of Defense, collecting and analyzing location-based data from satellites and sensors to support military operations and national security.

Brett Markham, NGA's deputy director, addressed the gap between public perception and technical reality during a keynote speech at the GEOINT Symposium on May 3. "There are certain people out there who want to know everything about everything all the time, 24/7, 365 days a year," Markham said. "In some circles, they think we have that ability today. I wish that were true."

That quote, reported by SpaceNews, captures the central tension driving NGA's AI strategy. The rapid adoption of AI-driven analytics has compressed expectations around how quickly intelligence can be delivered. But the technology hasn't caught up to the demand signal.

AI models now automate large portions of imagery analysis, detecting objects and flagging anomalies at scale. The physical reality of this work involves analysts staring at screens where algorithms highlight potential targets, reducing hours of manual review to minutes of verification. (The fatigue from scrolling through thousands of satellite images is real, and automation helps.) But algorithms still cannot interpret context the way human analysts do.

According to official documentation from NGA's website, the agency employs advanced analytics to decrease time needed to sift through staggering data volumes. This allows analysts to interact effectively with data, discover new objects, and ask intelligence questions that provide necessary context for decision-ready GEOINT.

The shift comes as NGA confronts a surge in data from satellites and other sensors. Over the next five to ten years, the agency projects a potential tripling of GEOINT data when considering all government and commercial satellite programs achieving full operational capability. That deluge forces changes in how intelligence gets processed.

Analysts are increasingly relying on AI "agents" to identify objects and surface unusual activity. This allows humans to focus on interpretation rather than initial detection. The workflow looks different now: instead of manually scanning imagery for hours, analysts receive prioritized alerts and spend their time understanding what those alerts mean.

A key area of development involves multimodal AI models. These systems combine multiple data types into a single analytic pipeline. In geospatial intelligence, that includes optical satellite imagery, synthetic aperture radar, infrared data, and non-imagery sources like text reports or signals metadata.

The approach is designed to increase analyst productivity. Optical imagery can be obscured by weather or darkness, while radar data produces a different representation of the same scene. By integrating inputs, multimodal systems maintain analytic continuity when one source is degraded.

Much of NGA's AI work takes place on classified systems, but the agency is increasingly dependent on commercial technology. "We have neither the time nor the expertise to build frontier AI models from scratch," Markham said, referring to leading-edge systems developed by a small group of companies.

The Pentagon has moved to formalize those ties. On May 1, it announced agreements with several major artificial intelligence firms, including OpenAI, Google, Nvidia, Microsoft, and Amazon Web Services, to deploy AI capabilities on classified Defense Department networks.

Within NGA, efforts are also underway to speed up acceptance of AI tools. The agency has launched a "computer vision model accreditation campaign," inviting companies to participate in a 90-day process to validate algorithms for national security applications. Firms with accredited models could gain an advantage in competing for government contracts.

These computer vision models are trained on satellite, aerial, and drone imagery. They're designed to do more than identify objects. By analyzing imagery tied to specific locations over time, they help analysts infer activity and track changes, forming the backbone of systems like the military's Maven Smart System.

Maven was established in 2017 as the Pentagon's flagship AI project to integrate AI into military workflows. The GEOINT aspects of Maven were entrusted to NGA in January 2023. Its state-of-the-art computer vision and AI capabilities are now integrated into various military analytic workflows to automatically detect, identify, characterize, extract, and attribute features and objects in imagery and video.

NGA Maven is already producing large volumes of computer vision detections for warfighter requirements across multiple operational locations. It has generated millions of data labels and lowered latency detections. Maven data and detections are fed into multiple other platforms and stands as an important thread in a tapestry of connected sensors from all branches of the armed forces.

The agency's stated objectives include utilizing high-quality computer vision that improves positive identification, geolocation, and speed. NGA will improve and scale existing overhead imagery Broad Area Search-Targeting and Full Motion Video Lines of Effort by employing novel algorithms and techniques.

Integration into the analytical workforce is another priority. NGA is striving to create common tooling, techniques, and standards by improving Structured Observation Management conversion to labels, increasing analyst feedback to models, and exploring synthetic learning and labeling.

AI reasoning and search features will be enabled by recent advancements in generative AI models. These will expand discovery of new types of objects. The goal is to assimilate AI into informed collection orchestration, where an established service continuously presents options for collection managers based on constellation constraints, standing needs, and dynamic events.

Implementation of an enterprise AI infrastructure rounds out the objectives. NGA will optimize imagery services, data storage, and data access, as well as compute to lower costs and increase speed. A common platform will incorporate data management for labels, models, and detections for analysts and customers.

The demand signal to know more with precision and timeliness will continue to grow, Markham noted. He pointed to rising expectations across military operations in space, air, maritime, and land domains. Rather than treating automation as a solution in itself, the agency is using it to reduce latency and narrow uncertainty for intelligence analysts.

"We're looking to automate and or apply artificial intelligence to certain workflows that get information from hours down to minutes in the hands of analysts, so that can be quickly turned to decision makers," Markham said.

Whether users actually pay for it remains the real question. The technology exists to process data faster, but the human element of interpretation cannot be fully automated. The gap between what stakeholders expect and what systems can deliver will likely persist for years.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn