The intelligence community has turned to a cutting-edge artificial intelligence product to help analysts hunt cyberattackers and understand leaks of classified information before Americans are victimized.
Officials at U.S.-based Primer say it is a point of pride for the company that its platform is used by analysts deep inside the National Security Agency and the company is ready to showcase the tools it put in the intelligence community’s arsenal.
NSA is America’s primary spy agency focused on collecting signals intelligence, the work of intercepting secret messages and breaking codes. The agency also conducts computer operations to protect against foreign threats.
Primer officials say their AI-powered platform enables analysts to spot cyber threat indicators across social media, track illicit finance and weapons, monitor adversaries’ tech capabilities, and uncover adversaries’ influence operations.
“What NSA analysts use it for is to augment the [cybersecurity signals] that they’re getting,” Primer Federal President Mark Brunner told The Washington Times. “We provide a different color to that, that helps them connect the dots as they’re trying to understand what’s happening in cybersecurity and any potential attack vectors.”
Primer has two main offerings, its Primer Command and Primer Delta platforms, which it hopes to use to build an all-source intelligence platform to give America’s spies a new kit of AI-powered tools.
As Mr. Brunner spoke with The Times at a Department of Defense AI symposium in Washington this month, Reddit posts fluttered vertically on a screen behind him in a social summary feed on Primer’s Command platform.
Command takes in open-source intelligence from around the world and aims to provide early-warning signals several hours sooner than traditional threat detection tools.
The demonstration of the platform displayed real-time posts from nearly two dozen social media sites and tens of thousands of news sources on a left-aligned panel, while an adjacent panel showed clues about where the monitored conversation was happening and who was involved.
Very recent history helps explain why the intelligence community wants such tools in its hands: Just one year ago, classified U.S. military and intelligence documents and images splashed across Discord, an online platform popular with gamers. Photos taken from above a suspected Chinese spy balloon that crossed the U.S., the Pentagon’s private assessment of Taiwan’s vulnerability to Chinese attackers, and more highly sensitive classified information made its way across social media platforms, from Telegram to X.
The federal government scrambled to understand where the classified information came from and how the material spread online. The alleged source of the disclosures, a low-level Massachusetts Air National Guardsman named Jack Teixeira, was arrested in April 2023 and U.S. analysts are looking to avoid a repeat.
Primer says it can help the intelligence community detect the spread of classified information and comprehend the information far more quickly than currently possible.
Primer CEO Sean Moriarty offered no guarantee that properly applying Primer would detect or understand the next Discord leaker, but he said, “You could certainly get there a whole heck of a lot faster.”
“The quicker we gain awareness the quicker we can act, which means you can contain the surface area of the damage and minimize it,” Mr. Moriarty said.
To drastically lower the time analysts spend looking for a needle in multiplying haystacks, Primer’s Demand product applies the company’s natural language processing models to consumers’ publicly available and classified information.
John Bohannon, Primer’s vice president of data science, said the powerful algorithms of “large language ” AI models represent a Rosetta Stone for the platform’s real-time data analysis across continents and languages.
Intelligence community analysts wanted the platform to answer complicated questions that could not be solved with the “Boolean” searches that people often perform on Google. An example of such questions includes identifying all of the arguments for and against an analyst’s hypothesis about any particular world event.
“We were stuck like everyone until basically multilingual embeddings and large language models that can actually reason through more subtle language cues came online,” Mr. Bohannon said.
More recently, Mr. Bohannon said, the company is hard at work improving the ability to detect an intent to harm, distinguishable from hyperbole, satire, or a range of other cultural expressions. The effort is a tall task for analysts looking to understand the difference between foreign disinformation campaigns and aggressive rhetoric that is offensive and protected speech.
Primer’s products are now commercially available, but the company has not forgotten about its roots with users in the Pentagon and intelligence community. Founded in 2015 and securing funding from the CIA-backed In-Q-Tel investment fund, Primer now has more than 130 employees and is experimenting with other AI tools for the U.S. military and intelligence services.
For example, Primer is working with an undisclosed intelligence community partner to develop the company’s audio-to-text technology for planes.
The NSA declined to comment for this article.