The Central Intelligence Agency (CIA) is entering the scale phase of its artificial intelligence (AI) implementation and has hundreds of AI projects underway, Israel Soong, deputy director of the CIA’s Office of AI, said Tuesday.
After years of work, Soong said during a Federal News Network webinar that the agency has successfully rolled out an AI literacy program to thousands of officers, built a state-of-the-art model repository, deployed a chatbot, and partnered with four major cloud providers to bring frontier model capabilities to the agency.
In some cases, Soong said that the agency is able to make new models available to users within seven to 10 days after they are released.
“Over the past few years, we’ve made significant investments in foundational infrastructure to make AI broadly available across the CIA,” Soong said. “And right now, we’re kind of at an inflection point. Like, we’ve moved decisively beyond asking, ‘Should we adopt AI?’ to asking, ‘How do we operationalize it responsibly at scale?’”
“In other words, we successfully completed an additional adoption phase, and now we’re expanding the capabilities and trying to transform mission execution,” he added.
The next step for the agency is to continue to accelerate the adoption of AI while eliminating redundancies and prioritizing resource allocations, Soong said.
“Now what we need is strategic stewardship,” he explained. “Now we’re trying to work AI into our mission workflows, not just as experimental tools, but as reliable capabilities changing how our workforce operates daily.”
One way the CIA is doing that is by embedding large language models, retrieval-augmented generation techniques, and chatbot-style interfaces directly into analysts’ daily workflows, Soong said. The goal, he added, is to free officers from time-consuming data review and allow them to focus on more complex mission tasks.
“Already, we’re seeing significant benefits to CIA’s intelligence mission,” Soong said. “It’s not the AI or the cloud in isolation, but all the elements of the AI tech stack working together, … that integrated approach of the two stacks is key.”
In the future, Soong says that AI agents will also play a role in the CIA’s daily operations, picturing a world where there are “teams of CIA officers managing teams of AI agents.”
“There’s a potential for AI to assist analysis, but also identify blind spots for the analyst and generate novel hypotheses for them to consider when they write about their analysis. Also, we could have AI autonomously identify intelligence gaps and generate novel collection strategies,” Soong explained, adding that the CIA is aiming to have officers spend 80% of their time thinking critically about intelligence instead of collecting it.
“Imagine this in the future,” he said. “AI agents have already … triaged the overnight intelligence for you. They’ve highlighted the most relevant operational developments, and they’re also combining and collating the intelligence reporting for you to review as you drink your morning coffee.”
For now, Soong said the CIA’s next AI push centers on deploying smaller models and AI agents at the tactical edge so officers in remote or hostile environments can use AI with little or no connectivity.