CIA looks to fast track AI adoption through cloud contract
imaginima/Getty Images
A CIA official said the agency is working with the five recipients of its Commercial Cloud Enterprise contract to “get commercial [large language models] up into our high-side environments.”
The CIA is pushing to quickly adopt artificial intelligence capabilities into its cloud environment but is cognizant of the computing power needed to support large language models, an agency official said on Wednesday.
During a webinar hosted by the Intelligence and National Security Alliance, Larry Taxson — chief of CLOUDworks in the CIA’s Directorate of Digital Innovation — said his office is working “to get modern capabilities in the hand of mission partners as soon as we can” through the agency’s recent cloud services vehicle.
The CIA previously awarded its Commercial Cloud Enterprise — or C2E — contract to five cloud service providers in 2020: Amazon Web Services, Microsoft, IBM, Google and Oracle. CLOUDworks helps intelligence community partners leverage commercial cloud services through the agency’s C2E contract, in addition to providing government hosting solutions.
Taxson said CLOUDworks’ main focus since the contract was awarded has been operationalizing C2E, which “means the ability for mission users to get access into all of our cloud service providers on multiple domains; so being able to bridge our networks, our accounting, our billing, our capabilities into those environments and then to start using them.”
He added, however, that the agency is also working closely with its five C2E cloud service providers “to see how we can get commercial [large language models] up into our high-side environments, hopefully by the end of this year or early next year.”
One of the early challenges with adopting AI capabilities across the public and private sectors has been allocating the computing power needed to support the tools. Taxson said “we have a lot of [graphics processing unit] availability within C2E today” but that “we're not going to suck the whole internet up.”
Some of the discussions around the CIA’s embrace of emerging capabilities moving forward have centered on the scope of the novel technologies that are really needed.
“Are we going to look at more SLMs, like small language models, that are going to be on niche topic areas?” Taxson asked. “What are we going to need to do that? Do I need GPUs or [tensor processing units] to train those?”
He added that the CIA is taking these factors into consideration as it works to build out its cloud environment through the C2E contract.
“The last thing we want to do is keep pushing our [cloud service providers] to give us all these capabilities and then they stay dormant that nobody uses them,” Taxson said.