Agencies Accelerate AI Acquisition as Frontier Models Evolve

Federal agencies are reshaping how they acquire, test and deploy AI as frontier models evolve on timelines measured in weeks or even days.  

The shift requires faster acquisition pathways, AI-ready data and infrastructure that can support increasingly autonomous systems while maintaining trust and human oversight, officials from the Defense Intelligence Agency, National Geospatial-Intelligence Agency and FBI said Wednesday at DoDIIS Worldwide in Tampa, Florida.

Federal agencies are shortening the time between the release of new AI models and their deployment to warfighters, intelligence analysts and law enforcement officers.   

At the DIA, officials have turned to other transaction authorities to accelerate acquisition including through its “NeedipeDIA” vehicle and a 501(c)(3) consortium, allowing the agency to acquire capabilities “in a matter of weeks, sometimes days” rather than months or years. 

“We knew we had to acquire and deliver quickly,” said DIA Chief AI Officer Maj. Gen. Robert Kinney. “For those who don’t know, TAs were started by NASA via the Space Act in the late ’50s to acquire technology quickly for the race to the moon, and DoD received those authorities in the late ’80s from Congress to do that, and we use them regularly for acquiring capabilities quickly.” 

At NGA, the focus is making the data AI-ready to remove potential bottlenecks before acquisition. CAIO Michelle Aten said the agency is working to establish common data and AI governance standards so that data does not become “the long pole of the tent” when new capabilities are deployed. She said they are also making sure that they are communicating those standards to agency partners to ensure interoperability. 

“We’re doing that so that we can make sure that when we do pursue rapid acquisition and rapid integration, rapid deployment of our AI capabilities, that the data is ready for us to be able to use those capabilities immediately,” she said.  

The FBI is pursuing a similar break from its traditional technology and contracting models and using a Broad Agency Announcement to identify innovative technologies and streamline procurement.  

Building Toward Agentic AI 

The agencies are also laying the infrastructure for a shift from standalone AI tools toward agentic systems that can connect data, applications and workflows. 

NGA is developing an agentic framework organized around individual tasks that subject matter experts have identified as priorities for automation. The goal is to avoid “running the same mile over and over again” by making agents broadly available and discoverable across the intelligence community. But trust must remain central, Aten said.  

“We need to ensure that we’re doing this in a trustworthy way … in a way that we have continuous monitoring of those agents as they’re deployed, so that if they, for example, exhibit any kind of anomalous behavior or aberrant behavior, that we can then pull them off the shelf, get them back to be retooled, and then redeploy them in a safe and responsible way,” she said.    

At DIA, the agency is on a “90-day sprint” to build an enterprise AI platform service and is developing Mission Model Context Protocol capabilities to provide a more universal way to access data. Kinney said the agency is “laying the foundation and the pipes” needed to build agents agents, possibly within the next year and, eventually, agents interacting with other agents.