Is America Ready for the Age of Cyber Warfare?

25:36See why Art News Magazine calls Independent 20th Century a staple of the season and experience historic modern art from around the world and one of New York's most iconic spaces, September 24th to 27th. To learn more, visit independenthq.com. This July, save$10 on an all-access subscription to Foreign Affairs. Get exclusive access to thought-provoking analysis by the world's leading policymakers,

26:18And now back to my conversation with Anne Neuberger. Your former boss at the NSA, General Paul Macazzoni, used to talk about persistent engagement. He, in fact, wrote a piece for foreign affairs in 2020, laying out some key elements of this doctrine. And one notable piece of it, at least as I read it from the outside, was the need to be much more aggressive in degrading adversary capabilities in cyber. Is the conclusion of that experience that we're less able to do that we might want because of the vulnerability on the defense side? Persistent engagement plays an important role. And in fact, I co-led, while at NSA, some of the first joint NSA Cyber Command Offensive Cyber Operations that really brought together the intelligence community and the military community to engage persistently, to use the intelligence we had about adversaries' infrastructure, their tools, the actors, to try to take them offline, make it harder for them to operate.

27:18I think what we learned when it came to the Chinese, and I should say, those efforts were successful and did indeed engage with and disable, take offline some adversary capabilities, expose, in some cases, the individual actors, in some cases, the companies that were part of China and I would say also Russia and Iran's campaigns. But I think what it also showed us was that the vast scale of China's capabilities, the number of units, the scale of their units, the number of companies that China uses in conducting its operations means persistent engagement to be really effective needs to be done at significant scale.

28:01And that was not the scale we were operating at, A. B, it also showed us that even if we try to throw sand in the gears of China's cyber operations, we still ran the risk because of the vulnerabilities of our own systems of that question I posed at the beginning, which is, do we have confidence that in a crisis or conflict, China couldn't disable or disrupt our military communications or critical parts of our infrastructure in order to delay military mobilization or put pressure on the homeland? And that is a key question that we need a more integrated approach across defense and offense, including persistent engagement, to be able to answer affirmatively.

28:43Which means if we go and try to take out some Chinese cyber unit that we are quite confident that they can't respond in a way that would be equally or more damaging? Agreed. And that brings us to the key question, which is to say, if we are setting clear lines vis-a-vis China, if we are saying that an attack that disrupts critical infrastructure, we will identify, we will attribute, and we will respond, we need to have that visibility across our defensive and offensive capabilities. That sounds a lot like deterrence, what we would traditionally call deterrence. I think there was a time as I read the debate about cyber strategy that there was a lot of talk of deterrence.

29:31And then people became more skeptical that the doctrine really applied in the cyberspace. You see an ability to kind of bring it back. Is that a kind of new development in thinking about deterrence in cyber? You know, the writing on deterrence, a number of the principles, I believe, do carry over into cyberspace. They're very different. For example, attribution is far harder in cyberspace, and it can take a bit of time, which can delay a response. But building the capability to attribute, and in fact, you may have seen over the last number of years, the number of U.S. and ally products that have been released attributing different activities to different actors, Russia, Iran, ransomware actors, China, has been to demonstrate that we have that capability and are continually strengthening to be able to attribute.

30:17because the difficulty of attribution has made cyberspace more of a free-for-all than a traditional kinetic space. So that first, we will attribute. The second, that we are resilient, that if an attack occurs, we will, of course, aim to prevent it, but at the very least, we will recover very quickly and that we will, and we maintain the capability and the will to retaliate. Those three parts, I think in deterrence language, it's, for example, deterrence by denial, that we are resilient, that we can recover, I believe do very much carry over from traditional deterrence theory, albeit with some changes and adaptations because cyberspace is different from the traditional kinetic space.

30:56Where are the other places where it diverges from the traditional view of deterrence? The number of actors and the availability of capability. So in the beginning, people would often compare nuclear deterrence and cyber deterrence. The difference is that there are cyber attack capabilities freely available online. In fact, we see criminal actors have been disrupting numerous American hospitals in the last couple of years, many given haven in Russia, for example. So there's more capabilities and more actors in this space, which does indeed sometimes make attribution harder and it makes enforcement far harder.

31:33But the attribution question has become easier in recent years than we thought it was, I don't know, seven or eight years ago when this seemed like a kind of insoluble problem? It is. And AI plays a key role in that and in the future as well. AI's ability to make sense of vast amounts of data, to, for example, trace the paths from an adversary's network through various covert infrastructure or compromised devices, they may stand up to hide their tracks all the way through to the target. AI has been playing a key role in noting the patterns of that and tracing that back as well. AI helps both offense and defense.

32:09It is a major help on one of the key problems in defense, which is collapsing the space between red and blue, as we say in military terms, between attack and defense, and to enable us to draw those links and identify who the ultimate attacker is. Which means, just to put a fine point on it, if we say that we think the Chinese government or the North Korean government or, for that matter, the Iranian government or Russian government is behind an attack, we now have more confidence than we might have a few years ago. We now can identify it in many cases, yes, and can do so faster and can have more confidence in our conclusions.

32:50I want to draw you out more on this question of how AI can enhance defense in this space and address some of those vulnerabilities that have been exposed by that Chinese campaign more broadly. The kind of fundamental asymmetry you focus on in the piece between an autocratic system and a democratic system where the U.S. government, at least at this point, does not force companies to put in certain defenses or reveal certain information as aggressively as it might and certainly as aggressively as the Chinese do. your answer to that is not to change the American system, but to use AI to kind of creatively work around that.

33:25How does AI help us address that asymmetry? Yes. I believe that AI can help us use our own system, our technological and innovation leadership. America is a leader in the world and within our system of democracy, where there is private sector ownership of critical infrastructure to still have visibility and be able to prevent attacks from being successful. You know, one of the hardest questions we have in national cyber defense is the question of where should we spend our cybersecurity dollars given the vast scale of America's critical infrastructure and given the degree of what the community calls technical debt, old systems.

34:05You know, I recall visiting a Louisiana water system with President Biden to view the rebuilding. And it was still, in many cases, a very old system. Some parts of it were manual and analog. In some ways, that's easier. The systems weren't yet connected to the Internet. But that kind of old systems that have been connected to the Internet while not being secured are across U.S. industry. They're the reason, for example, we've seen so many hospital disruptions. So where does the private sector and government spend its scarce cybersecurity dollars to really buy down risk? And the challenge we have is that even though in the intelligence community, we may often have insights about adversary attack capabilities or planning, there's a limited degree to which we can test those out against live systems.

34:50Nobody wants to risk running a Russia or China's war plan or cyber attack capabilities that we may uncover against the U.S. energy grid for risk of actually causing negative impact. So digital twins, AI allows us to create virtual replicas. In the corporate world, those have been used for quite some time. Rolls-Royce creates a digital replica of its engines in order to test different safety scenarios. You'll have companies that take manufacturing processes and create digital replicas in order to figure out where do you optimize it without taking those manufacturing processes offline. We can do the same with our most critical infrastructure, creating, for example, a virtual replica of a part of our energy grid in order to model different cyber attack capabilities.

35:34and frankly, model resilience responses to determine which investments we should make and how secure and resilient our grid actually is against real attack. So to sum up, I would say today too much. We have a U.S. government and intelligence community and cyber and other intelligence experts on one hand, passing information to private sector owners and operators, but they're not working together on the same system. They don't have a copy of an important critical infrastructure system to actually model attacks and resilience and ensure that we can make the right investments. We're having this conversation at a time when the Trump administration is blowing through all kinds of longstanding norms about the relationship between the state and the private sector and demanding investment in information and concessions from private companies that would have seemed, I think, fairly shocking not long ago.

36:29So given perhaps some of these shifting norms without suggesting that that's the right approach, do you see more need for a kind of fundamental change in the balance between the private sector and the government, whether it's requiring certain investments in cybersecurity that are now kind of a patchwork driven by the bottom line and by financial considerations, as you note, or demanding more information sharing or something else? Absolutely. The risk has only grown. AI helps on offense significantly in terms of finding vulnerabilities and more rapidly building exploits. One of the benefits we've had in cybersecurity is that it's truly been a bipartisan field.

37:06Republicans and Democrats come together to say, ensuring that the nation's secrets can be protected digitally, ensuring that people's communications can be protected has really been a bipartisan goal. And in fact, you know, as we've seen, whether it's here or whether it's in 5G policy, policies have remained pretty much, you know, President Trump put in place certain 5G policies to start countering Chinese dominance via Huawei. President Biden continued those. President Biden's cyber executive orders. President Trump continues those. So I think that bipartisanship is our strength. In the last administration, President Biden overturned decades of bipartisan Democratic and Republican resistance to having minimum cybersecurity requirements.

37:50And President Biden put those in place for ports, for pipelines, for airports. I recall, you know, I would have a spreadsheet of every airport in the country, every pipeline in the country provided to me by the regulatory agency that rated them red, yellow or green. and the plan the agency had as a regulator to work with those companies to get them to yellow and to get them to green. And significant progress was made in the last four years. That's foundational. The second step now is to take that work deep in the partnership between America's AI companies, the technological leaders, the companies that own and operate the grid, that own and operate other key important sectors are telecoms, going back to our earlier conversation, and government intelligence cyber experts bring them together to work together on digital replicas of that critical infrastructure to secure it to the level needed, to use our AI expertise for national cyber defense.

38:46As you look with the last few years of experience at the ways in which AI enhances offensive capabilities and enhances defense, how do you think it changes the offense-defense balance in this space? It's a great question. It changes the equation if we use AI for cyber defense aggressively to find and close vulnerabilities. AI will help both. As I mentioned earlier, AI helps find vulnerabilities. The second step determines whether one's an attacker or defender. If one finds vulnerabilities in order to exploit them, get a foothold into a network either to collect intelligence or to stay there or hide your tracks for a later period in time to disrupt the operations potentially of that network, then one is an attacker.

39:34If one uses AI to find vulnerabilities to then close them, then one is a defender. And in fact, during the last demonstration, we launched an innovative program with DARPA, which is essentially DoD's R &D agency, to really challenge the AI companies to use AI to find and close vulnerabilities in open source code. And indeed, the program was really successful. In year one, they found nearly all of the pre-placed vulnerabilities and were also able to identify significant patches in the last year that went even further. So there's massive promise in this space on cyber defense. So I understand the ways in which these AI tools help defenders, but it does seem that there are plenty of ways that offensive capabilities have surely gotten much better in the last few years and ways that should, I think, scare us a bit.

40:24Yes. Indeed, you know, the purpose of the piece was a call to action because if we do not move fast, artificial intelligence will only accelerate China's advantages, will only accelerate the advantages for offense and China's ability to use its great firewall to integrate with AI to find and block attacks even more effectively. We truly believe the U.S. possesses the technical capabilities, the innovation, and the work across government and private sector to reclaim the advantage in the digital battle space. And this is really a call to action to say, we can't wait any longer. We need to act now.

41:02Your time at the NSC and the National Security Council did not just involve working on cyber issues. It also involved emerging tech more broadly. Emerging tech has, I think, become one of the central arenas, if not the central arena of competition between the US and China and more broadly at this point. So I think it's worth zooming out and looking at the state of that competition more generally. I think starting with AI, you wrote a piece, I think in the last week or so of your time in government earlier this year about the ways in which AI is changing intelligence more broadly, not just in the cyberspace.

41:34A lot of the focus here has been on the ways in which this has made surveillance much more powerful and ubiquitous. That makes human intelligence much harder. It's harder to be a covert operative when you're being surveilled all the time. And AI enabled facial recognition and gate analysis technologies and everything else enable governments to identify who you are. But as you look at the kind of transformation of intelligence more broadly. How do you see those effects sort of shaking out and what really strikes you about the way this changes the way intelligence works? You know, that piece that I wrote in Foreign Affairs actually started as a classified piece a year before to outline for the intelligence community the promise and the significant peril and risk to intelligence operations from AI.

42:19And the reason that is, is that people design intelligence operations. And as such, there's a predictability and a pattern to them, whether they're signal intelligence operations, human intelligence operations, or imagery collection, you know, when satellites move over particular places. And AI's strength in amassing and making sense of vast amounts of information and seeing patterns in them is a real risk to our intelligence operations. So red teaming with AI our own operations is needed to ensure we find those patterns and predictability and fix them first so that it doesn't lead to uncovering operations.

42:55On the flip side, AI's abilities similarly can help us deliver faster, better, and more focused intelligence for policymakers and for the nation's security. And there are a number of exciting pilots happening across the intelligence community and changes to incorporate AI into operations. As you look forward three or four or five years, do you think human intelligence, by which we mean the kind of traditional thing we think of as spying with operatives kind of going to other countries and trying to get information from government officials, trying to steal secrets, do you think that will become all but impossible just given the power of those surveillance technologies?

43:36I think I'll need to dramatically change. The rise of deep fakes. You may have seen, for example, a company called Eleven Labs can now take voice and change it, multiple languages, different genders, different personas easily. So that will make human operations different. So the challenge really for us is to reimagine how we do intelligence in the age of AI. One good example, you know, actually two good examples, I would say. Today, the intelligence community is very much organized as a collect and collect by different domains, images, signals, human interactions, process, analyze, and report.

44:19And that process has different controls built in to ensure that it's accurate, to ensure, for example, there's multiple sources for particularly sensitive insights. But it is a process with very distinct steps. When you think about where we are today, if one can deploy an AI model into when we're doing satellite collection to identify at that point, you know, based on the shape, based on the behaviors, for example, of ships in the Indo-Pacific or different platforms in the Indo-Pacific, a threat or an opportunity, we need to ask ourselves, what's the fastest, most effective way to get those insights to the consumer?

44:56And perhaps the traditional set of steps and sets of integrity validation need to change. They need to be updated while still feeling confident that the outcome of that is correct. AI is making everything move more quickly. It collapsed the decision cycles in the kinetic space, in cyberspace. So I think going back to the drawing board to say, with the power of AI to make sense of massive amounts of information, to add precision at the point of collection, to translate at scale, how do we ensure that our intelligence community makes the most of that to protect the nation and inform policymakers rapidly on a broad scale of topics?

45:38You noted that a lot of these efforts have seen bipartisan support, have been bipartisan efforts over the last several years. I think to some degree that has changed, at least in certain ways, in the last several months since Trump has come back into office. You've seen lots of firings within the intelligence community, including the head of NSA for reasons that I think are still not entirely clear what the substance was, other than the fact that Laura Loomer, the right-wing activist, had focused on him. And he may have once worked with General Mark Milley, the former chair of the Joint Chiefs whom Trump hates.

46:07And your old boss, Paul Nuccasone, is not someone especially popular in Trump circles because of work he did on, you know, kind of election integrity. Does that kind of politicization worry you? You've seen pretty stark warnings from senior former intelligence leaders, including Mike Hayden, who was head of NSA and CIA under George W. Bush, so not a partisan exactly, warning that this could lead to just a degradation in American intelligence capabilities and the risk of a real intelligence failure that we could see in the coming years. How do you assess those risks? You know, I served in the intelligence community for almost 15 years, coming in at mid-level and then rising to be deputy head of our global intelligence operations and standing up our cybersecurity mission.

46:49And perhaps in many ways, most importantly for your question, Dan, serving as NSA's first chief risk officer, rebuilding internal controls and operations following the Snowden media leaks of 2013. And from that time inside the halls, I grew to deeply, deeply respect the intelligence community, intelligence professionals who often work very long hours in unglamorous spaces where their successes are secret and their failures are publicly known. And there very much is a culture of non-partisanship. People keep politics out of the workplace because the nation's security is not political. and we feel a commitment in the intelligence community when we take an oath to the Constitution.

47:32And when that oath is renewed, that's who we're protecting, America and the American dream. And I would say, in fact, that one of the most moving ceremonies that I've ever observed is within NSA, they're both military and civilian professionals. About 40 % of NSA, I believe, is military. When you walk in to at least the old building I worked in for many years, as you walk in, there is a black granite wall, and it's engraved with names, some of which still say anonymous. And at the top of the wall, it would say they served in silence. And during my time at the agency, a number of names were added to that wall, engraved by hand.

48:13And when family members of the individual who fell came, professionals, military, and civilian would line up on the whole so that the family would see the respect everybody had for the individual who had fell in the line of duty. That's who the intelligence community is. And I think, you know, full stop, I also recall my own experience serving as chief risk officer post-Snowden. And it was a deeply disconcerting time. And I think what I found so disconcerting was that inside NSA, people genuinely felt, you know, we're here to protect America. Of course, we live by American values and laws. And outside NSA, when I would be talking with broader American public or with American companies, that was severely questioned.

49:02And I think the gap really was within the walls of the intelligence community, because in a democracy, it's on the intelligence community to, while keeping its operations in the shadow, know, talk enough about its goals, its values, its people so that the average American trusts what's going on inside those walls. And I think when we look at our current time, clearly that trust doesn't exist across some swaths of America. So some deep reckoning and reflecting on how to rebuild that to ensure our military and the intelligence community particularly continues to play that role in America is very, very much needed.

49:38I mean, I'm struck though that you have senior intelligence community officials being removed from their positions because they've given answers to questions that parts of the Trump administration don't like. Surely that has some effect on the quality of analysis that gets to policymakers. It's just hard to imagine that doesn't have some detrimental effect. It is deeply concerning. It's deeply concerning. You know, the intelligence community's policies and processes are meant to ensure that doesn't impact analysis. And I no longer am in the intelligence community. So anything I would say would be a hypothetical, but I would hope and trust that the culture of the community, the sense of its past and its future, would continue to ensure intelligence professionals are doing their work to ensure that what policymakers need to know, they know, and it's unaffected by the processes around them.

50:29That's obviously very, very difficult, though. But it's something that I would hope when people reflect on their oath to the Constitution, they reflect that that's who they're doing their work for. I think the other interesting difference between the Trump and Biden administrations on this issue has to do with kind of theory of emerging tech more broadly. I think some of this you can chalk up to idiosyncratic factors and probably some private sector influence and corruption. But there is, I think, also a different underlying theory when it comes to how the U.S. should win AI competition, what AI competition means.

51:06You had the Biden administration much more focused on export controls and keeping some of our technology out of Chinese hands, especially the AI diffusion rules that was fairly restrictive about what kinds of AI tools could be shared with others around the world. I think the Trump view is, you know, we should let American companies make as much money as possible. And that's the best way of staying ahead. How do you understand those differences and how do you assess the kind of state of that competition at this point as you look at the changes in U.S. policy? Export controls and ensuring that America's edge in chips are not available to our adversaries to promote their military and intelligence capabilities because AI's power is so transformative is a fundamental part of that competition.

51:50But it's only a part. The real way you win a competition is by running faster, not by holding a competitor down. So I think as important as export controls are, what's as important is America's adoption of AI in our economy, in the transformative fields like drug discovery, in our national security, in all the ways we talked about producing more, faster, and better intelligence by adopting AI and streamlining processes to fully take advantage of AI. incorporating AI in our military operations, for example, so that we can detect adversary platforms that may have been covert until now. That's the way that we will truly win across our economy and across our national security while still doing our best to ensure that our most capable technologies don't fall into the hands of our adversaries.

52:37Of course, that's important, but we need both components for success. Where, as you look at our attempts to run faster, do you think we're doing well and where do you think we're not doing enough? We have a lot of exciting pilots. I think what's important is that we scale those and scale us to transform our operations. So for example, you know, one could see where in the intelligence community, there are tens of thousands of intelligence reports that were produced over the last 10 years on various topics. Imagine training a set of specific models across those intelligence reports to enable policymakers and analysts to engage as we do with chat GPT, to ensure that we're getting the benefit of all of that insight and using it rapidly and effectively.

53:22Those are places where we can move faster to ensure that we're using the promise of AI, both for our economy and our national security. To a degree that I think would have surprised both of us a year ago, the Trump administration has taken aim at some of the pillars of the American innovation ecosystem, as we talk about it, some of the drivers of American success in emerging tech, whether you look at attacks on universities or really making it much harder for both researchers in academic context and at companies to bring in foreign talent and work with foreign talent, big cuts to research budgets at NASA and other parts of the US government, obviously all the tumult at the FDA and the CDC and other parts of our kind of health infrastructure.

54:04What of those changes worries you if they're sustained? What would really kind of undermine the ability of the United States to stay ahead in some of these emerging tech races, as you put it? America is a technology and innovation leader globally. That's, of course, due to our dynamic and entrepreneurial private sector. It's also due to government funding of R &D, particularly R &D that's foundational science. It isn't necessarily applied to existing problems. That certainly is a source of our innovation. and something that we know that we need to continue. And I think now is really the opportunity to consider what are the fields of R &D that are most changing and where is it most important for programs to focus, particularly given that in the area of AI, so many of the advancements are coming from the private sector.

54:53So in many ways, government focus in those areas need to complement the advancements that are happening in the private sector today. And that is a good note to end on. Thank you so much for doing this today. And thanks for the series of great pieces you've written for Foreign Affairs in the past several months. We will look forward to more in the months ahead. Thank you so much for having me, Dan. And thank you for the role Foreign Affairs serves in talking about difficult issues in a way that people around the world can learn, reflect on, and act. So truly appreciate the time with you today.

55:30Thank you for listening. You can find the articles that we discussed on today's show at foreignaffairs.com. The Foreign Affairs Interview is produced by Kanishk Theroux, Molly McEnany, Ben Metzner, Caroline Wilcox, and Ashley Wood with audio help from Todd Yeager. Our theme music was written and performed by Robin Hilton. Special thanks as well to Arena Hogan. Make sure you subscribe to the show wherever you listen to podcasts. And if you like what you've heard, please take a minute to rate and review it. We release a new show every Thursday. Thanks Thanks again for tuning in.

56:28Free weekly newsletter featuring more of the clear headed analysis that you enjoy here on the podcast. Sign up today at foreign affairs dot com slash spotlight.