Adopt the Adversarial Point of View:' Cybersecurity Lessons From a National Intelligence Technology Leader

The world of intelligence gathering is shrouded in secrecy. Intelligence agencies around the world use clandestine operations and classified technology to support everything from military planning to counterterrorism to cyberdefense.

Amit Meltzer spent three decades working in Israel's national security apparatus, including as chief technologist for Mossad, the country's famed national intelligence agency. “The main reason I left was that my wife said: 'Enough!'”

Since his retirement from government in 2010, Meltzer has spent most of his time consulting companies on cyberdefense and advising venture capital firms and cybersecurity startups, including Recorded Future. On a Friday morning last month, I caught up with Meltzer to talk about his time working in national intelligence and how it shaped his thinking on cybersecurity. The conversation below has been lightly edited for length and clarity.

The Record: Thank you for taking the time to talk.

Amit Meltzer: I don’t normally do interviews, but it has been a decade since I’ve retired from public service, it's history more than news.

TR: What can you tell me about your background in intelligence?

AM: I didn’t grow up in intelligence—I was a combat soldier when I did my normal service back in the late '70s. I got into national intelligence in a meandering roundabout way, first as a security guard when I was a student, and then I got into technology. I’m not by nature an intelligence guy—I’m a techie.

Up until 20 to 25 years ago when the explosion of digital communications and digital devices started changing the nature of the world, of course it was changing the nature of national intelligence and gathering information. Knowledge of certain areas of technology became a Class A intelligence asset. I was never an expert in anything, because I moved laterally in my career across six or seven domains. I became sort of a hub for composite knowledge that became important for crafting new intelligence support systems and analysis systems and later on intelligence support for operations. I sort of moved with the flow, you might say, moving from run-of-the-mill IT and engineering into intelligence analysis and launching data science analytic teams to evaluate mission-critical data. Later on, I was the architect of the past generation of intelligence gathering, analysis, dissemination, and operational support systems for the organization. In essence I kept pushing technology to serve more functions within the intelligence landscape, because it’s anything but monolithic: There are dozens of different teams and specialized units doing specific intelligence gathering or special operations, and they each have different demands.

After several years of developing such specialized systems, I was part of the drive to integrate everything and create a fusion system, which became the cornerstone of intelligence dissemination. Once you get a fusion system and realize how much raw intelligence you can get from the data, it changes the way you evaluate the intelligence both for targeted intelligence gathering and operational support. You get a lot more planes of reference for how information is relevant to operations. If you showed how field operations are run today to someone who was a spy in World War II or the Cold War, it would look like a Marvel movie to them. Both the field team and the back office is connected—the number of devices and tools keeps expanding. It’s not really Star Wars, but it’s highly technical now, and the amount of communications devices, covert devices, monitoring devices, and command and control features you have now is really a game changer compared to what any intelligence agency was doing two decades ago. I was at the edge of this when it exploded, and it was an exhilarating experience.

TR: What have you been up to since leaving the government?

AM: About 50% of my time I spend consulting corporations on cyber resilience and defenses. And about 40% to 50% of my time I do investment due diligence—I’ve evaluated about 300 cybersecurity startups over the past 11 years. When I see a startup I like, I give it an assessment that will help it get seed money but also get directly involved, either as a mentor or an advisor. I’m usually on three or four advisory boards at any given time, and I never take money because they’re startups and strapped for resources, so I only take options. One-in-eight in general makes money and is successful—it’s mostly something I do for fun, and to support the innovators. Founders are young and bright-eyed and energetic and know nothing about commercialization. They will fall flat on their faces if they’re not coached directly into the market, because ideas don’t sell—products sell.

Because of the coronavirus, I essentially closed all my consultations in North America, India, and Southeast Asia. If I can’t fly, I can’t consult on cyberdefense strategy via Zoom or Signal. They’re too paranoid, probably for a good reason. So I closed everything and switched to VC investing. We’ll see what happens once we get back to normalcy though.

TR: You mentioned how you were in charge of building and designing a fusion system when you were in government, which is kind of a precursor to Recorded Future. Where did the idea come from?

AM: It essentially came from screw-ups. With two or three near-disasters, when we did the post-mortem we discovered we had the pertinent information in other sources from other feeds from other teams in the organization that weren’t disseminated fast enough or at all. There’s a strong compartmentalization in such agencies by design, so information doesn’t flow laterally—it’s only utilized by the one who got it. It’s not useful for intelligence or operations after a certain point. We had to create a very complex fusion program, and naturally intelligence and security operations guys are not very... community oriented. I’m trying to be extra polite, but what I’m trying to say is that they don’t like to share. These guys are amazing, they have amazing skills, but they don’t play well together. So we had to figure out a way to create a fusion platform for intelligence while maintaining the compartmentalization and the rights to access so there won’t be a breach of operational secrecy and compromised assets, which is the greatest wrong you can do in such organizations.

The layout was very complex. We did a multi-level stripping of the data so we could fuze it together. It sounds easy, but it’s problematic because in the end you get really complex… hundreds of different access groups and combinations to enforce the compartmentalization while allowing the underlying algorithms and search engines to actually access the data. If it was too sensitive, you would get a message saying you need to call the guy in the other unit and ask for access rights. In essence, it was extremely complex at the intelligence level but was much easier at the raw data level, because all you had to do was create security for the source materials. And the source is very sensitive—one thing you don’t want to do in intelligence is compromise the source, and it happened several times, but not on my watch. The one thing you don’t want is to give someone a report, and the report circulates all the way to the prime minister, who says, “Oh, we had such a great tip about Hezbollah from Adam Janofsky,” and boom, the source is gone because they were compromised. So source protection was the paramount directive.

The other challenges were things that are now common, but were pretty cutting-edge 20 years ago. How do you create a data model that allows you to ingest billions of data records a day from hundreds of sources that still makes sense? The data was messy—when you get raw data from clandestine operations, it makes no sense and doesn’t have any structure. We had thousands of data models that you had to arrange into a single coherent landscape without oversimplifying the model. You can’t just pour everything into an elastic search engine and hit search. You get limited functionality and results if you simplify everything down to blocks of text. 

TR: How did you overcome these challenges?

AM: It was a challenge in engineering and data modeling, and I’m not shy so I took advice from everyone I could corner—including people from sister agencies, academia, and leading technology vendors. And if I couldn’t corner them, I would use the police to make fake charges and arrest them so I could interrogate them in prison—I’m joking, of course! But I took advice from everyone I could, and eventually we created a fusion database that became the cornerstone of emerging intelligence and analytics. It was very far from perfect, but was a good first step, and the people who took over after me made much better systems built on our initial effort.

I created several teams and mentored them into a functioning department. Since I’m a mediocre mathematician—my fifth grade math teacher let me know it—I recruited mentors from academics. I had at least two senior professors at any given time to mentor and coach the data science teams, and it was a good decision because it created a rapport and interaction that supported breakthroughs. Those teams had an excellent track record and won national security awards. 

TR: How did your experience in intelligence help you with cybersecurity?

AM: There are several connecting points, some of them may be roundabout. For example, with intelligence you have to adopt the adversarial point of view, because in many cases you want to be able to get some intelligence and you don’t have access. The way to get the data is to make it seem like a good idea for the target to give you the data—this is a common operational model. In technology it’s more complex. You have to formulate an adversarial view and think what are the areas I can get the data from, and then evaluate the entire environment of the target and figure out what normal operation they will carry out that will give you the data that you want. A good example is the SolarWinds attack. It was highly professional and utilized an existing, widely-used update mechanism. The attackers did a very good leap-frogging of that channel to infiltrate hundreds of organizations probably and install their monitoring trojans or whatever you want to call them. You have to figure out how to get a handle on it—an adversary penetrates using covert methods.

When I switched into cybersecurity, I used the same way of thinking to evaluate the security setup and find the weak links in the operation that, once utilized, can create a breach. In essence, I was thinking as an attacker and overlaying it on the existing infrastructure and suggesting how organizations can create additional monitoring points or alerts so they can have a fast enough reaction to a roundabout attack aimed at breaching their systems.

I think all governments are the same, from Putin to Biden to Netanyahu. All governments are bureaucracies. They don’t move fast—they are slow by design, and it’s not a bad thing. Governments that move faster than its citizens usually fall."

— Amit Meltzer, former CTO in the Israeli intelligence community

The other big thing is understanding the two inherent weaknesses in the entire sphere of technology—the people and the code. It boils down to who are the most dangerous people in an organization, and 99% of the time it’s the coders and the infrastructure support team. These guys have no security sense and there’s very little punishment if they make horrendous errors that lead to breaches, and they do it every day. It takes a lot more time and effort to be a safe coder. And the most common and easiest way to initiate a breach is by fooling people with spam, phishing, and other methods to insert the initial payload. Intelligence will teach you that everything is about trickery, and you have to focus on the assets that you never want certain things to happen to. 

TR: What do you think the role of government is when it comes to technology? Should they be early adopters?

AM: Well I worked for the government, and I think all governments are the same, from Putin to Biden to Netanyahu. All governments are bureaucracies. They don’t move fast—they are slow by design, and it’s not a bad thing. Governments that move faster than its citizens usually fall. The only scenario where governments should move fast is war. 

The challenge with infrastructure is how does a government or an agency adapt its procedures and norms to utilize new technologies. Consider the paradigm shift when cellular came along, all of a sudden a lot of tactical communications from the field to the control room switched from finding payphones like Superman to having instant messages. It went through a dramatic change. And organizations are resistant to change, people don’t like change, and government organizations are the hardest to change. The only mitigating factor is that operational command is more flexible than a bureaucracy. If you go to the CIA they will be more open to adapt their operational control methods than the Department of Agriculture or Energy, because they have constantly shifting goals unlike other agencies.

If you want to create a fast response team in government, you can’t do it even within an agency—you have to create a new organization that’s focused on fast response. The Israeli Army did that several years ago because they had like 17,000 special forces teams all over the place with no one talking to each other. If you want to move as fast as the attackers, create your own JSOC environment. Cyber Command is a behemoth bureaucratic agency, and it’s moving as fast as the DEA or ATF because it’s too big and mostly focused on crafting better defenses. That’s not a bad goal, but in the shifting challenges of cyberdefense, if you want a fast reacting team or a preemption team or a proactive counterattack team you can’t do it in government—you have to go outside of existing agencies and establish an operation center. 

TR: In what other ways do you think governments should change their approach to cybersecurity?

AM: Attacks by organized crime or nation states are at least as effective and professional as the people who are trying to defend, and sadly there are people in the IT security community who still think of hackers as a 17-year-old pimpled guy in the basement of his mom’s home writing code in Python. If you look at the attack on SolarWinds, you need a highly skilled, organized, and operationally-centered infrastructure at APT29 [Editor’s note: Some early reports suggested that APT29, a Russian-linked threat group, was responsible for the attack. The U.S. government has said it is likely Russia was behind the incident, but has not blamed a specific group yet.] to actually create effective harvesting. The initial breach is one achievement, but creating a working campaign to harvest all the available data you have access to in hundreds of organizations is an operational platform—it’s not two guys in a basement, it’s more like 40 guys working shifts around the clock in a well-structured and well-maintained campus. These people are professionals and you have to give them respect. So in essence the government has to sponsor similar operations for cyber defense, and if they want to own it they have to offshore it to another organization.

The other thing they can do is incentivize people in the private sector, which moves much faster than the government... I’m guessing Putin has a very good incentive plan for cybersecurity specialists to work for the government, otherwise they would run to Romania and go into cybercrime. A good incentive plan can get the government the speed and rate of innovation that they can’t do on their own. But an incentive plan must be managed to leverage innovation into both government-centric and private sector oriented defense posture.

Old scores take a long time to settle in the Middle East—it’s an area with long memories. There are certain areas I won’t ever fly to.”

— Amit Meltzer, former CTO in the Israeli intelligence community

TR: With what you were saying about governments moving fast in times of war, I was reading an article in the Jerusalem Post from a few years ago about how you used AI against Hezbollah. Does that fit with what you’re saying?

AM: First of all, I don’t believe in AI—I’d rather use the term machine language. AI is a marketing label, and I’ve yet to see a working AI. But machine learning has become crucial for intelligence analytics, and effectively integrating it into intelligence work is the goal of any army or security agency on the globe. 

When you think about what does an army do against a guerilla operation that hides their operations and is focused on asymmetric warfare, you have to think about how you can use intelligence to level the playing field. There are two things any clandestine organization wants to keep at all costs. One is the actual whereabouts of their assets—the bunkers, missiles, and what have you—and the other is the command and control infrastructure. Even if it’s highly compartmentalized, you still need a command structure, and if you can derail that and gather intelligence about the majority of the physical locations of the assets of your enemy, in this case Hezbollah, then intelligence can be transformed into an attack plan.... If you can monitor and disrupt the command and control infrastructure, you essentially take out their ability to act as a cohesive unit, because they’re left with people on bikes and scooters to pass on commands. The same goes for cyber. If you want to disrupt a cyber campaign, you need to understand how it’s run and how you can disrupt it. 

TR: When you were in government, were you ever concerned you would be personally targeted by Hezbollah or a hacker?

AM: It’s been a decade, so I guess I’m not relevant now for the knowledge I have. But old scores take a long time to settle in the Middle East—it’s an area with long memories. There are certain areas I won’t ever fly to, like the Persian Gulf. Even if they let me fly—there are peace agreements between Israel and the United Arab Emirates, Bahrain, Morocco, Sudan—I won’t go there, because I don’t need the off chance that I won’t get back. My wife seems to want me around. About a quarter of the world I won’t fly to. It’s not a high-level threat, but just a general precaution. 

TR: What are you most proud of from your time in government?

AM: I talked about the fusion center—I put a lot of effort into that, and the effort was also to overcome pushbacks. There are always pushbacks, and something to understand about national security agencies is that many if not most of the senior managers have past experience in field operations. These people are the scariest people in the world, and not because they’re like James Bond—no one carries a gun anymore in these agencies. These people have interpersonal skills at a level that you can only imagine, and their craft is to make other people do what they want and think it’s their own idea. Working with them is scary. They’re amazingly persuasive, they get their way, they’re overly aggressive. Pushing your agenda in this kind of environment requires a lot of effort and tenacity. I pushed the fusion center into the first stage, and I’m guessing and hoping it’s at a much better stage, but I’m very proud of the initial effort because it wasn’t a given. It went against the grain. 

The second thing is extending the intelligence field into open source. I helped set up open source intelligence gathering operations when it wasn’t being considered. It’s a very complex operation, especially because cyberspace is not uniform, and when you think about intelligence gathering the obvious is crawling across websites and gather information, and later crawl across social media and get the data on the guys you are after. They learned how to be cautious very fast. But there are two gifts that keep on giving: The first is the dark net, the unregistered portion of the internet, and if you got there early enough and established your presence in these ultra-paranoid areas, you could get information about the activities of many perpetrators. It can take years to a decade to establish your presence in such an environment. The second, which is more technically tricky, is the deep web. Consider repositories maintained by government entities, public entities, academic entities, guilds… they have a wealth of data on each one of us. I created a map of what repositories I wanted to get from a single country, and I came up with about 500, and 50 were critical. Getting those repositories is difficult, and buying them out is very expensive, and they don’t like people coming in and harvesting them. They put up countermeasures, and will cut you off and blacklist you immediately. You need to find workarounds... It’s highly complicated, but extremely fruitful.