[WAPO] Aaron Brown was working as a CIA case officer in 2018 when he wrote a post for an agency blog warning about what he called "gait recognition." He cautioned his fellow officers that computer algorithms would soon be able to identify people not just by their faces, or fingerprints, or DNA — but by the unique ways they walked.
Many of his colleagues, trained in the traditional arts of disguise and concealment, were skeptical. One called it "threat porn." But Brown’s forecast was chillingly accurate. A study published in May reported that a model called FarSight, using gait, body and face recognition, was 83 percent accurate in verifying an individual at up to 1,000 meters, and was 65 percent accurate even when the face was obscured. "It’s hard to overstate how powerful that is," Brown said.
Brown’s story illustrates a profound transformation that is taking place in the world of intelligence. For spies, there is literally no place to hide. Millions of cameras around the world record every movement and catalogue it forever. Every action leaves digital tracks that can be studied and linked with others. Your cellphone and social media accounts tell the world precisely who and where you are.
Further, attempts at concealment can backfire in the digital age. An intelligence source told me that the CIA gave burner phones to a network of spies in a Middle Eastern country more than a decade ago and instructed them to turn the phones on only when sending operational messages. But the local security service had devised an algorithm that could identify "anomalous" phones that were used infrequently. The network was exposed by its attempt at secrecy.
"The more you try to hide, the more you stand out," Brown explained. He wouldn’t discuss the Middle East case or any other operational details. But the lesson is obvious: If you don’t have a cellphone or a social media profile these days, that could signal you’re a spy or criminal who’s trying to stay off the grid.
Brown, a wiry former Army Ranger and CIA counterterrorism officer, is one of a small group of ex-spies who are trying to reinvent American intelligence to survive in this age of "ubiquitous technical surveillance," or UTS. He launched a new company this year called Lumbra. Its goal is to build AI "agents" that can find and assess — and act upon — data that reveals an adversary’s intentions.
Lumbra is one of nearly a dozen start-ups that I’ve examined over the past several months to explore where intelligence is headed in 2025. It’s a dazzling world of new technology. One company uses data to identify researchers who may have connections to Chinese intelligence. Another interrogates big data systems the way an advertising company might, to identify patterns through what its founder calls "ADINT." A third uses a technology it calls "Obscura" to bounce cellphone signals among different accounts so they can’t be identified or intercepted.
Most of these intelligence entrepreneurs are former CIA or military officers. They share a fear that the intelligence community isn’t adapting fast enough to the new world of espionage. "Technologically, the agency can feel like a sarcophagus when you see everything that’s happening outside," worries Edward Bogan, a former CIA officer. He now works with a nonprofit called 2430 Group — the number was an early CIA cover address in Washington — that tries to help technology companies protect their work from adversaries.
The Trump administration recognizes this intelligence revolution, at least in principle. CIA Director John Ratcliffe said during confirmation hearings he wants to ramp up covert operations, with officers "going places no one else can go and doing things no one else can do." That’s a commendable goal, but if the agency doesn’t reinvent its tradecraft, Ratcliffe’s bold talk may well fail. Traditional operations will only expose the CIA and its sources to greater risk.
A CIA spokesperson said this week in response to a query: "Today’s digital environment poses as many opportunities as it does challenges. We’re an adaptable agency, and it is well within the ingenuity and creativity of our officers to develop ways to navigate effectively in complex environments. In fact, we are exploiting many of the same technologies to recruit spies and steal information."
Brown takes hope from the work that younger CIA officers are doing to reimagine the spy business: "Some of the agency’s smartest people are working on these tradecraft problems from sunup to sundown, and they are coming up with unique solutions."
The CIA’s technology challenge is a little-noted example of a transformation that’s happening in every area of defense and security. Today, smart machines can outwit humans. I’ve written about the algorithm war that has revolutionized the battlefield in Ukraine, where no soldier is safe from drones and precision-guided missiles. We’ve just seen a similar demonstration of precision targeting in Israel’s war against Iran. For soldiers and spies everywhere, following the old rules can get you killed.
(Illustration by Raven Jiang/For The Washington Post)
The art of espionage is thousands of years old. The Bible speaks of it, as do ancient Greek, Persian and Chinese texts. Through the ages, it has been based on two pillars: Spies operate in secret, masking who they are and what they’re doing (call it "cover"), and they use techniques to hide their movements and communications (call it "tradecraft"). Modern technology has shattered both pillars.
To recall the mystique of the CIA’s old-school tradecraft, consider Antonio J. Mendez, the agency’s chief of disguise in the 1980s. He described in a memoir how he created ingenious facial masks and other deceptions that could make someone appear to be a different race, gender, height and profile. Some of the disguises you see on "The Americans" or "Mission Impossible" use techniques developed by Mendez and his colleagues.
The CIA’s disguises and forgeries back then were like works of fine art. But the agency in its first few decades was also a technology pioneer — innovating on spy planes, satellite surveillance, battery technology and covert communications. Its tech breakthroughs were mostly secret systems, designed and built in-house.
The Silicon Valley tech revolution shattered the agency’s innovation model. Private companies began driving change and government labs were lagging.
Seeing the disconnect, CIA Director George Tenet in 1999 launched the agency’s own venture capital firm called "In-Q-Tel" to connect with tech start-ups that had fresh ideas that could help the agency. In-Q-Tel’s first CEO was Gilman Louie, who had previously been a video game designer. In-Q-Tel made some smart early investments, including in the software company Palantir and the weapons innovator Anduril.
But the CIA’s early attempts to create new tradecraft sometimes backfired. To cite one particularly disastrous example: The agency developed what seemed an ingenious method to communicate with its agents overseas using internet addresses that appeared to be news or hobby sites. Examples included an Iranian soccer site, a Rasta music page and a site for Star Wars fans, and dozens more, according to investigations by Yahoo News and Reuters.
The danger was that if one agent was caught, the technology trick could be exposed — endangering scores of other agents. It was like mailing secret letters that could be traced to the same postbox — a mistake the CIA had made with Iran years before.
Iran identified the internet ruse and began taking apart CIA networks around 2010. China soon did the same thing. The agency’s networks in both countries were largely destroyed from 2010 to 2012.
In a 2012 speech during his stint as CIA director, Gen. David H. Petraeus warned that the fundamentals of spying had changed: "We have to rethink our notions of identity and secrecy. ... Every byte left behind reveals information about location, habits, and, by extrapolation, intent and probable behavior."
But machines moved faster than humans in the spy world. That’s what I learned in my weeks of on-the-record discussions with former CIA officers working to develop the espionage tools of the future. They describe a cascade of commercial innovations — instant search, mobile phones, cheap cameras, limitless accessible data — that came so quickly the CIA simply couldn’t adapt at the speed of change.
Duyane Norman was one of the CIA officers who tried to move the system. In 2014, he returned from overseas to take a senior operations job. The agency was struggling then to recover from the collapse of its networks in Iran and China, and the fallout from Edward Snowden’s revelation of CIA and NSA secrets. Norman remembers thinking that "the foundations of our tradecraft were being disrupted," and the agency needed to respond.
Norman convinced his superiors that in his next overseas assignment, he should try to create what came to be called "the station of the future," which would test new digital technology and ideas that could improve offensive and defensive operations. This experiment had some successes, he told me, in combating surveillance and dropping outmoded practices. But the idea of a "station," usually based in an embassy, was still a confining box.
"You’re the CEO of Kodak," Norman says he warned Director Gina Haspel when he retired in 2019, recalling the camera and film company that dominated the industry before the advent of digital photography. Kodak missed the chance to change, and the world passed it by.
When I asked Norman to explain the CIA’s resistance to change, he offered another analogy. "If Henry Ford had gone to transportation customers and asked what they wanted, they would have said ’faster horses.’
"That’s what the CIA has been trying to build. Faster horses."
The intelligence community’s problem was partly that it didn’t trust technology that hadn’t been created by the government’s own secret agencies.
Mike Yeagley, a data scientist who runs a company called cohort.ID, discovered that in 2016 when he was working with commercial mobile phone location data. His business involved selling advertisers the data generated by phone apps. As a cellphone user moves from work to home — visiting friends, stores, doctors and every other destination — his device reveals his interests and likely buying habits.
Yeagley happened to be studying refugee problems back then, and he wondered if he could find data that might be useful to NGOs that wanted to help Syrians fleeing the civil war into Turkey. He bought Syrian cellphone data — cheap, because it had few commercial applications. Then, on a whim, he began looking for devices that dwelled near Fort Bragg, North Carolina — where America’s most secret Special Operations forces are based — and later appeared in Syria.
And guess what? He found a cluster of Fort Bragg phones pinging around an abandoned Lafarge cement plant in the northeast Syrian desert.
Bingo! The cement factory was the headquarters of the Joint Special Operations Command task force that was running America’s war against the Islamic State. It was supposed to be one of the most secret locations on the planet. When I visited several times over the past decade as an embedded journalist, I wasn’t allowed to walk more than 50 yards without an escort. And there it was, lighting up a grid on a commercial advertising data app.
Yeagley shared that information with the military back in 2016 — and they quickly tightened phone security. Commanders assumed that Yeagley must have hacked or intercepted this sensitive data.
"I bought it," Yeagley told them. Even the military’s security experts didn’t seem to realize that mobile phones had created a gold mine of information that was being plundered by advertisers but largely ignored by the government.
Thanks to advice from Yeagley and many other experts, data analytics is now a growing source of intelligence. Yeagley calls it "ADINT," because it uses techniques developed by the advertising industry. Who would have imagined that ad salespeople could move faster than secret warriors?
(Illustration by Raven Jiang/For The Washington Post)
Glenn Chafetz had been station chief in three countries when he returned to Langley in 2018 to take an assignment as the first "Chief of Tradecraft" in the operations directorate. It was the agency’s latest attempt to adapt to the new world, succeeding the Ubiquitous Technical Surveillance Working Group, which in turn had replaced the CCTV Working Group.
"People realized that the problem wasn’t just cameras, but payment systems, mobile apps, WiFi hubs — any technology that produced data that lived permanently," Chafetz recalls. But there was still a lack of understanding and resistance from many officers who had joined the CIA when there were no cellphones, digital cameras or Google.
For the older generation, tradecraft meant executing "surveillance detection routes" to expose and evade trackers. Case officers had all gone through field training to practice how to detect surveillance and abort agent meetings that might be compromised. They met their assets only if they were sure they were "black," meaning unobserved. But when cameras were everywhere, recording everything, such certainty was impossible.
Chafetz lead a team that tried to modernize tradecraft until he retired in 2019. But he remembers that an instructor in the agency’s training program admonished him, "New officers still need to learn the basics." The instructor didn’t seem to understand that the "basics" could compromise operations.
The tradecraft problem wasn’t just pervasive surveillance, but the fact that data existed forever. In the old days, explains Chafetz, "If you didn’t get caught red-handed, you didn’t get caught." But now, hidden cameras could monitor a case officer’s meandering route to a dead drop site and his location, long before and after. His asset might collect the drop a week later, but his movements would be recorded, before and after, too. Patterns of travel and behavior could be tracked and analyzed for telltale anomalies. Even when spies weren’t caught red-handed, they might be caught.
The CIA’s default answer to tradecraft problems, for decades, was greater reliance on "nonofficial cover" officers, known as NOCs. They could pose as bankers or business consultants, say, rather than as staffers in U.S. embassies. But NOCs became easier to spot, too, in the age of social media and forever-data. They couldn’t just drop into a cover job. They needed an authentic digital history including things like a "LinkedIn" profile that had no gaps and would never change.
For some younger CIA officers, there was a fear that human espionage might be nearly impossible. The "station of the future" hadn’t transformed operations. "Cover" was threadbare. Secret communications links had been cracked. The skeptics worried that the CIA model was irreparably broken.
After all my conversations with veteran CIA officers, I’ve concluded that the agency needs an entirely new tool kit. Younger officers inside recognize that change is necessary. Pushing this transformation from the outside are scores of tech-savvy officers who have recently left the CIA or the military. It’s impossible at this stage to know how many of these ventures will prove successful or important; some won’t pan out. The point is the urgent need to innovate.
Let’s start with cellular communications. That’s a special worry after Chinese intelligence penetrated deep inside the major U.S. telecommunications companies using a state-sponsored hacking group known as "Salt Typhoon." A solution is offered by a company called Cape, which sells customers, in and out of government, a mobile network that can disappear from the normal cellular grid and protect against other vulnerabilities.
Cape was founded in 2022 by John Doyle, who served as a U.S. Army Special Forces sergeant from 2003 to 2008 and then worked for Palantir. His "Obscura" technology bounces mobile phone identifiers among thousands of customers so it’s impossible to trace any of them. He calls his tactic "opportunistic obfuscation."
One of the most intriguing private intelligence companies is Strider Technologies, founded in 2019 by twin brothers Greg and Eric Levesque and chief data officer Mike Brown. They hired two prominent former CIA officers: Cooper Wimmer, who served in Athens, Vienna, Baghdad and Peshawar, and other locations; and Mark Pascale, a former station chief in both Moscow and Beijing. The company also recruited David Vigneault, former head of Canadian intelligence.
Strider describes itself as a "modern-day economic security agency." To help customers secure their innovation and talent, it plucks the secrets of adversaries like China and Russia that steal U.S. commercial information. China is vulnerable because it has big open-source databases of its own, which are hard to protect.
Using this data, Strider can analyze Chinese organizations and their employees; it can study Chinese research data, and how it was obtained and shared; it can analyze the "Thousand Talents" programs China uses to lure foreigners; it can track the contacts made by those researchers, at home and abroad; and it can identify connections with known Chinese intelligence organizations or front companies.
Eric Levesque explained to me how Strider’s system works. Imagine that a software engineer is applying to work for an international IT company. The engineer received a PhD from a leading American university. What research did he conduct there? Was it shared with Chinese organizations? What research papers has he published? Who in China has read or cited them? What Chinese companies (or front companies) has he worked for? Has this prospective employee touched any branch of the Chinese civil-military conglomerate?
Strider can operate inside what China calls the "Great Firewall" that supposedly protects its data. I didn’t believe this was possible until Levesque gave me a demonstration. On his computer screen, I could see the links, from a researcher in the West, to a "Thousand Talents" program, to a Ministry of State Security front company. It turns out that China hasn’t encrypted much of its data — because the authorities want to spy on their own citizens. China is now restricting more data, but Levesque says Strider hasn’t lost its access.
We’ve entered a new era where AI models are smarter than human beings. Can they also be better spies? That’s the conundrum that creative AI companies are exploring.
Scale AI sells a product called "Donovan," named after the godfather of the CIA, William J. "Wild Bill" Donovan. The product can "dig into all available data to rapidly identify trends, insights, and anomalies," says the company’s website. Alexandr Wang, the company’s founding CEO (who was just poached by Meta), explains AI’s potential impact by quoting J. Robert Oppenheimer’s statement that nuclear weapons produced "a change in the nature of the world."
Vannevar Labs, another recent start-up, is creating tools to "influence adversary behavior and achieve strategic outcomes." Its website explains: "We develop sophisticated collection, obfuscation, and ML (machine learning) techniques to provide assured access to mission relevant data."
The company’s name evokes Vannevar Bush, an MIT engineer who headed the U.S. Office of Scientific Research and Development, which oversaw all major U.S. research projects during World War II, including the launch of the Manhattan Project.
Lumbra.ai, the company launched in March by Brown, seeks to create what he describes as a "central nervous system" that will connect the superintelligence of future AI models with software "agents." After leaving the CIA in 2021, Brown met with Sam Altman, the founder of Open AI, to refine his thinking. To describe what agentic AI can do, he offers this hypothetical: "We can find every AI researcher, read all the papers they’ve ever written, and analyze any threats their research may pose for the United States." Human spies could never be so adept.
LUMBRA
"No one said we have to collect intelligence only from humans," Brown tells me. "When a leader makes a decision, someone in the system has to take a step that’s observable in the data we can collect." Brown’s AI agents will create a plan and then build and use tools that can gather the observable information.
Brown imagines what he calls a "Case Officer in a Box." Conceptually, it would be a miniaturized version of an agentic system running a large language model, like Anthropic’s Claude. As an offline device, it could be carried in a backpack by anyone and left anywhere. It would speak every language and know every fact ever published. It could converse with an agent, asking questions that elicit essential information.
"Did you work in the Iranian weaponization program?" our Case Officer in a Box might ask a hypothetical Iranian recruit. "Where was your lab? In the Shariati complex? Okay, then, was it in the Shahid Karimi building or the Imam Khomeini building? Did you work on neutron triggers for a bomb? How close to completion was your research? Where did you last see the prototype neutron triggers? Show me on a map, please."
The digital case officer will make a great movie, but it’s probably unrealistic. "No one is going to put their life in the hands of a bot," cautioned Wimmer, a fabled CIA recruiter. The agent would suspect that the AI system was really a trick by his own country’s spies. Brown agrees that recruiting a human spy will probably always require another human being who can build the necessary bond of trust. But once that bond is achieved, he believes technology will enhance a spy’s impact in astonishing ways.
Here’s the final, essential point. Human spies in the field will become rare. Occasionally, a piece of information will be so precious that the CIA will risk the life of one of its officers, and the life of an agent, to collect the intelligence in person. But that kind of face-to-face spying will be the exception. The future of espionage is written in zeros and ones. The CIA will survive as a powerful spy agency only if it makes a paradigm shift.
|
...sees all, knows all, gets 'em all in the end...
for espionage.