How to Win in Cyberspace: A Conversation with Anne Neuberger | 27 November 2025

Fantastic to be here. Thank you.

Absolutely. So it's really great to be here with all of you. It's great to be a fellow here at RUSI. I've long admired the work that the team here does, so it's nice to have the chance to also see it from the inside. And Salt Typhoon and Vault Typhoon are actually names that private cyber companies came up with to refer to two particularly long-running, intrusive Chinese cyber programs. Salt Typhoon was the extensive compromise of telecoms, large, medium and small, in the US, in the Middle East, in Asia. And clearly everything rides on telecoms, right, from corporate IP to our private conversations to national secrets. So really an interesting source of intelligence. China's compromise gave them the ability to essentially make a copy of any call that they wanted, collecting large amounts of information. And because cell towers track locations as well, track phones of interest.

But more on the espionage side, Vault Typhoon was a distinction in that that is China prepositioning in critical infrastructure, power, pipelines, water systems. Those don't have real intelligence value, unlike telecoms. So our community assessed that that was because China was prepositioning to disrupt the operations of these critical systems during a potential period of crisis or conflict, either to put pressure on a particular government. I'm talking about America now, but we saw this prepositioning in countries around the world during a crisis or conflict, or to delay, for example, military mobilization during a conflict, because like here, you know, the US does not have distinct, you know, our soldiers, sailors and service members deploy from regular airports, regular ports. So that would be the purpose of those. So two broad Chinese campaigns with different goals, but significant national security implications.

That's exactly the distinction, because when we used to think about, you know, the nation's secrets, we have images of filing cabinets in a military or in an intelligence agency, you know, and a spy breaking in in order to somehow get out a few documents. Today, the scale of information, and frankly the fact that they ride on telecom, the fact that China was able to turn, you know, major telecommunications companies in America essentially into an intelligence tool, there are national security implications. And that's the point the article I make, which is in democracies like the US, like the UK, we carefully prescribe the role of government in defending or protecting private companies for the simple reason that cybersecurity rides on monitoring a network, and we've had concerns, obviously, about democracies monitoring a network, monitoring potentially private communications. So as such, we've carefully prescribed the role. We share information regarding a threat. We in the US, we started putting in place minimum cybersecurity requirements for key critical infrastructure because we saw that they were still being compromised by pretty basic things. There's new, I believe, draft legislation in the UK that would propose to do the same. But we don't actually put our best government hackers to defend those networks because, as democracies, we say that crosses the traditional public-private boundaries. When we look at Salt or Vault, I think the average American citizen would say, 'What do you mean the Chinese could collect a conversation at will? How could that be?' And with Vault, 'What do you mean China's prepositioned in a water system and could disrupt it? I expect clean water. I expect my government to be ensuring that's the case.' And yet...

That's not the case. So, you know, and if we cannot defend, as we all know, deterring an adversary, a capable adversary requires three things. You can credibly convey there will be consequences for actions that you say is unacceptable. And then you signal and you convey those consequences. Well, given that the government is not in a position to be confident about the defense of those critical infrastructure, that then has knock-on effects in terms of what we can deter. So I talk about ways that we can potentially change that, because fundamentally defense and offense are linked in cyberspace.

So they caused significant concern because of the implications we just talked about. So in the case of Salt Typhoon, essentially, you know, the prior administration overturned two decades of really bipartisan policy to not regulate private sector cybersecurity. The sense was we'll do more information sharing. We'll talk about the threat, that will incentivize the action to, you know, to use better multi-factor authentication, to actually encrypt data, to do all the steps we've been talking about consistently. But I think in both these incidents, in both these extended compromises, we saw that in many cases those basic practices weren't in place. So that led the Biden administration to say we can no longer continue and to radically change the approach to actually put in place regulations for pipelines, airports, railways, etc. That was a first step, and indeed the telecoms regulations had not been put in place until Salt Typhoon was uncovered. When it was, you know, the Biden administration regulator for telecoms did announce regulations in that space. The Trump administration has rolled those back. So we're back to where we were. But for the other sectors, those remain in place. So that was the first part, the defense part. The second part was to say, well, how do we ensure that we act more seriously in this space? So we engaged the Chinese government, conveyed to them that these activities were unacceptable, shared technical details to show what this was. But fundamentally, the deterrence question of how do you convey consequences if activity is seriously of concern is one where a new approach is needed.

Yes. So China is an authoritarian country, and in China, significant parts of critical infrastructure are state-owned enterprises. So first, as an authoritarian country, China monitors all of its domestic networks. Right? We know the Great Firewall is a system of monitoring. The purpose of that is both monitoring to censor communications, but it also puts China in the position to be monitoring networks to detect threats and block them potentially. As we think about cyberspace, you know, there's no one strike that deters, that conveys a message, right? Instead, it's much as traditional warfare. It's a back and forth. It's who gains the upper hand, who retains the upper hand. And as such, the asymmetry between how democracies can defend their critical infrastructure and how authoritarian countries, notably China, can defend their critical infrastructure ends up restraining how we'll use offensive capabilities. You know, I think about, you know, in the run-up to the war in Ukraine, we had now as public sensitive intelligence that Russia intended to invade Ukraine. And at the time, we had meetings in the Situation Room to explore. We knew that the US government, the UK, our allies were putting together a rigorous sanctions package to put pressure on Russia. Both we conveyed in anticipation to say, 'Don't do this. There's going to be significant economic costs,' but also to start ratcheting up the pressure to try to contain Russian action, right? And one of the first questions was, how might Russia respond? Well, one way we know Russia has a sophisticated, excellent offensive cyber program, and we were concerned about that. And so the first question when asked about, you know, how if Russia chooses to retaliate via cyber, how confident are we that we can defend critical infrastructure? The reality is because of the posture we have for those sectors, when we know we have minimum regulatory, we at least know their basic posture, but many don't. And as a result, the government's best hackers and defenders are not in line to parry, which ends up constraining what we'll do.

So I would say that, as you noted, first, there were a series of cyber attacks. Let's talk about them because I think they could have been more impactful. Let's talk about that. But second, I think our community's general takeaway was that, as we all know, Russia expected the invasion and overtaking of Ukraine to be, well, not quite a walk in the park, but to be far easier, to be relatively days. Exactly. Right. So as such, the extended offensive cyber campaign was not actually built out, and the information in the Russian system about who knew about the invasion was very much contained. And then what we saw was, and this is always a central tension between intelligence and offensive cyber, once the invasion started to struggle, there's always a decision between when the Russians compromise the system, do they seek to gain intelligence? So, for example, if you're compromising Ukraine's military planning system, do you want to learn everything you can on what they're planning to do, or do you want to brick the thing? And that's a strategic decision. And I think our sense was that as the conflict continued and the struggle, the intelligence value exceeded it. Now, with that, the Ukrainians do get some credit, and an unlikely entity, Elon Musk, gets some credit. So the evening before Russia's invasion, exactly as you know me, Gordon, right, Russia conducts a disruptive cyber attack to essentially knock out Ukraine's military communication system, right? The system that soldiers use in the field to talk securely. And talking securely in the field matters because clearly if you're two units separated where you're saying, 'Hey, I'm going to go left, you know, because or we see a potential ambush on the right,' you need those communications to be secure, otherwise your adversary will pick them up in a heartbeat. So Russia disrupts Ukraine's military communication system the night before. As Gordon noted, it was an American company providing the service. Well, that meant that when Russia invaded, Ukraine did not have a secure military communication system. Elon Musk stepped up and he provided Starlink. So that's an example of where, you know, less that if not for that, there would have been more impact from that. And then we did start to see Ukrainian-Russian wiper attacks on Ukrainian systems. I had called the head of a number of private sector companies before the invasion and said we were very concerned about a NotPetya scenario, a Russian destructive cyber attack that just spreads. And I'd called and said if you guys see anything destructive that's spreading, you know, call. And indeed we got a phone call and ended up connecting those companies with NATO's team and with teams across European governments, but also those particular companies were partnering with Ukraine, and Ukraine had given them permission to just shut down anything that they see. So as a result, you know, those companies essentially shut down those attacks very quickly and effectively. And finally, Ukraine moved a lot of their systems to the cloud, assisted by European and American companies, which added resilience and made it harder to disrupt them as well. So I think it's a combination of things that meant that Russia's offensive cyber program, while it played a role, it didn't have the impact they anticipated.

And that's why there are so many interesting lessons learned.

My own feeling is that the balance is in defense because I'll explain why. You know, offense, an offensive actor needs to get in one way. You look around a room like this, windows, doors, right? They just need to find one where there's a crack and get in to achieve their goal. A defender has to be monitoring each one and knowing that each one is safe, right? So as such, the space that you're monitoring, and everybody here who's done defense knows there's lots of manual work. You're checking, is there a misconfiguration? Is there a patch that's just come out? Right? Like, so AI gives a lot of ability because you can, you know, gather information, prioritize more quickly. I think it helps a defender more with their harder goal versus on offense. Now, both goals start the same with AI. Whether you're offending, whether you're a hacker or whether you're a defender, your first step is finding a vulnerability. And that clearly AI does in both cases. The second piece is then if you're a hacker, you want to build an exploit. If you're a defender, you want to build a patch. And AI coding capabilities now makes those faster too. So it's really on us as people with regard to can we get the defensive AI community to roll up its sleeves and get to work in that way. And I think, you know, quite frankly, the reason I suggest digital twins in the article is one of the hard problems we have in cybersecurity is that let's say an intelligence agency learns about the latest offensive technique from North Korea, from Iran. Because we're not monitoring many of those systems, those domestic systems, we actually don't know, will this work? What impact would it have if it was used? So in many cases, the first time we see what impact is...

Precisely, and we can all agree that's not so. The idea of digital twins is, and they're heavily used in manufacturing, to essentially put sensors on systems and then create a virtual replica of a system. And when you have a virtual replica, you can model. So if you've picked up, you know, the offensive cyber technique, or you've actually gotten the code, if you're lucky, you can actually model that on a digital replica and say, would it work? And if so, what do we need to do to make sure in practice it would fail? And I think that's what's so powerful because we don't risk that on a live system because you don't want to cause the impact and you don't quite know what will happen. A replica lets you do that. So I think it can have a powerful impact because in both the UK and the US we have tons of old systems, a lot of technical debt. We know we need to spend money on cybersecurity, but we've got to get the highest return on investment for the money we spend. So by actually exercising or simulating attack techniques on virtual replicas of systems, it could potentially give us the way to build the most effective ROI cyber defense plan and also better prepare for what we know are high-end adversary capabilities.

So I don't think, in my own view, I don't view it as a race of whoever gets over the finish line is winner takes all, because I think AI has broad implications for our economies and our national security. And in many cases, I think there's three questions we ask about who ends up with the upper hand, but it really doesn't matter if one is the upper hand and one is ten feet below, right, using a metaphor. And I think the three questions we ask ourselves are, first, in the places where AI is a dramatic game-changer, so let's say in intelligence, where too much data or the scale of data is a weakness, not a strength, so AI helps make sense of that, and it lowers the bar for what used to be massive investments of compute and of skill to do that, right? So I think the first question is, in intelligence and national security, who adopts it faster in places where AI makes a difference? And I use intelligence as one example, and in that case, it's a never-ending, there's no race, it's continuously, and adoption is hard in all of our government systems, right? I often think that the countries that do best on those are those in tough neighborhoods who are small with a technical talent base. Sometimes big isn't a strength. B is what parts of the economy have jobs that will be replaced, and what's going to happen to those people? Because displaced workers, a loss of dignity, a loss of purpose is not a good thing in an economy. And a big part of our national security strength is our economic strength. Right? So China is now converting its strength in manufacturing to robotics. It can train AI models based on the knowledge it has on manufacturing processes, and it can train AI-driven robots in a way that for the US, at least, which has lost a lot of our manufacturing, we really need to consider how we effectively compete in robotics and to ensure that sectors where we do have some strength, like medicine, surgery-driven robots, we move quickly to retain that. That's the second space. And then I think the third is whose technology does the world run on, because AI will be such a game-changer. And in that case, it's often, as you know so well, as we all know, it's not necessarily the best mousetrap, it's the cheapest and good enough for most cases, right? So that's the way I personally look at the competition with China rather than a race, because I think it's going to be an ongoing competition. And to me, at least, those are the questions we should be asking all of our countries in terms of how will we do.

Absolutely. So first I would say there's regulations and then there's the cultural piece. You know, in our intelligence agencies, for example, we've been doing machine learning on translation, as you would expect, right, for a decade. And yet it took many, many years till analysts were willing to adopt it because they were worried they'd lose their jobs. And that's in an agency where there's so much data and where humans, you want your people looking at stuff that's been triaged by a machine to identify it's worth their time. Right? So I think when we think about constraints, regulation is not the only constraint. It's also the cultural uptake. I'll give another great example. China's ports are largely automated operations, robots moving heavy loads. In the United States, our longshoremen, our port workers struck, you know, did a strike last year around Christmas time, biggest retail season, right? So between Thanksgiving and Christmas. What was the reason for the strike? To gain assurances that automation and AI would not be deployed in those ports in order to save jobs, right? And, you know, when I saw that, I found it quite sad because what would have, under the fear of losing jobs, is understandable, but one would think you'd want folks saying, we want guaranteed job training for how we train robots, for how we use robots, for how we learn how to code these systems because they're always going to need to change, right? We want to move to the future, but we need help and we want to be guaranteed that help along the way. So I think we focus on regulations, but there's at the very least that cultural piece as well. But clearly an approach that guarantees the safety we're incorporating AI would have significant risks is very much needed. And at least what I saw from a cybersecurity perspective was leveraging regulators who knew the sectors to say what kind of potential new regulations are needed or what updates are needed was a more effective way to make it purpose-built to like what are new risks.

You know, so it's a great question. I would think that, you know, probably both of our mothers when we were younger said, 'Before you take on the bully on the playground, make sure you know what you're trying to accomplish.'

She didn't quite say that to me, but I think, you know, I know what you mean.

But, and I think that's the core question we have to ask ourselves. Like, you know, it sounds easy to say we need to go on offense. Okay, what are you trying to accomplish? And then what does that really mean? So if we're worried about prepositioning in a water system, okay, what does offense mean? Are you going to preposition in their water system? Under the laws of armed conflict, we don't view that as a valid target. Okay. Then what are you going to go after? Their military systems. Okay. Like, but that is where the work has to be done. And then it's what do you convey? And then when you convey to them you're there because you want to deter them, you want to go on offense, right? What if they just find it and kick you out? What have you accomplished? Right. So that level of detailed work in our offensive cyber operations. Let's learn. But starting with what's your strategic goal? Will you get it done that way?