Tech Is Security: Anne Neuberger, General Partner, Head of Global Affairs, a16z | Betting on America

Absolutely. So as you said, a lot of these technologies are fundamentally dual use. You know, one of the things I was fortunate is when I served at the National Security Agency, I had the opportunity to both lead our offensive mission in terms of intelligence collection and also then subsequently lead our defensive mission in terms of protecting the security of nuclear command and control systems, weapon systems, etc. And working, leading amazing teams that did that. And then moving over to the White House to look at the policy side again of both offense and defense. So that's the way we need to look at these technologies, and as you said, not only each of them independently, but increasingly how they bleed together. For example, when we think about AI in the context of cyber, AI's ability to find vulnerabilities quickly also comes together with generative code execution, which those two, finding a vulnerability and then exploiting it at scale, is what is transforming AI-driven cyber attacks. On the flip side, as you know, whether on offense or defense, the first step is finding a vulnerability. An attacker exploits it, a defender patches it. Thinking about how do we bleed that together on the defensive side to catch up with an attacker is one of the most challenging, interesting areas because I actually think the delta of impact is far more in defense because you have distributed sensors, massive amount of information. The economics of cybersecurity has been broken forever, because there's, we have large amounts of tech debt in this country, and the ability to figure out what do we patch first, how do we do it at scale, what's the cascading impacts, is something you can now do with AI. So back to your point regarding dual use capabilities and the link between, the bridge between the national security and the tech world. I think as a country we both want to say we must be leaders in this technology. We want the world to run on a Western stack, and yet we also want to protect our own systems against adversarial use of these technologies, whether it's, like I said, in AI and cyber, whether it's in AI and synthetic bio, or whether it's in quantum, where there's so much exciting happening.

Absolutely. You know, so quantum is a space that still generates a lot of skepticism, understandably, because it's been, you know, viewed as a science project for so long or a development just over the horizon. And there are a number of areas. When we say quantum, I'll break them out into three. One, which is the most mature and advancing, is quantum sensing. You know, quantum essentially relies on the magnetic, on the forces at the center of the earth. So you can do very precision time and location. Certainly in an era of GPS denial. We first experienced it when the Russians were operating out of Syria. Commercial aircraft are now experiencing it in various areas around the world. It is really important for whether it's a commercial pilot or whether it is a precision weapon to know where it is and to know where its target is. Quantum sensors have made significant advancements to where they can help do that, and the systems have gotten a lot smaller. That's exciting. It's also a dual use capability because there are some concerns that it can illuminate, for example, where submarines may be because of the same precision location at the earth's core. So that's one good example. A second area of quantum, which is the one that people in the, you know, digital resilience world think about with great fear, is a potentially cryptographically relevant quantum computer, a computer that can break one of the two core types of cryptography. Why does cryptography matter? Because it's the root of all digital security. When you and I go shopping on Amazon, we see a padlock in the browser. That padlock represents a series of systems behind it that validate who we are, who the shopping site is, and ensures that many people can shop on many sites safely. That entire ecosystem is potentially at risk. Would there be a quantum computer that could break that core cryptography? So there have been some advancements in the number of qubits needed to do that, and as a result you've seen companies like Google bring forward the date to which they're moving to these new algorithms. Many large digital infrastructure companies have already moved. So when you and I are texting on Apple, they're already, one of the algorithms they have in Apple, for example, on our smartphones, are a quantum-resistant new cryptography algorithm. But that's an area where now companies...

Exactly. No, thank you. Exactly. So you now see, I get a lot of phone calls from people saying, okay, is this thing real among all the other threats? What should I be doing? And my answer always is, it's real. We don't know when, it'll be likely within the next five years. So the wise thing to do is to start doing an inventory. Figure out where in your organization what kinds of cryptography you have developed. Where do you have data that's still going to matter in a few years, right? Where do you have, like for example, if you're a pharma company, your R&D, you know, where do you have systems where the security really matters. Get that sense so that you can start transitioning to new algorithms. Some of them still have hiccups. They have large key sizes. They can be slower. Use and broader use will fix that, but starting that transition is wise. And then the final part of quantum is actual quantum computing. And I think that's the area where there's often the most questions. And you have companies doing real work, starting to build the algorithms in medicine, in banking, to where there's a belief that simulation and certain kinds of computations quantum computers can do uniquely well, potentially together with today's regular classical computers. But in the area of maturity, that's what we see. So it's a space where developments are happening, but there's probably still some fair amount of skepticism as well.

Well, you know, it's so interesting you're asking that, Naveen, because I think in many ways that was the core foundation of our debates in the Biden administration regarding artificial intelligence policy. Some of us had really watched the 5G case and seen how Western technologies were largely upended by a very careful Chinese playbook, to where, you know, in the 5G case, China initially stole the technology from leading American companies, then started developing it on its own, scaled 5G across China, increasingly closing the domestic market to outsiders, which enabled it to really build the tech much more cheaply and also get just that scaled experience in telecom, which was at the root of now China's productivity buildout, and then heavily subsidized exports abroad because China understood that from a national security perspective everything rides on telecom, right? Your national security secrets, your IP secrets, your government's communicating with its citizens, to where by the time, with credit to the first Trump administration, we woke up as a country and the first Trump administration started sounding the bell: this is dangerous. It's dangerous to have everything riding on a core Chinese tech. And I can tell you as somebody technical, it is near impossible to secure a telecom system running with Huawei's tech. Why? There are legitimate reasons for backdoors, e.g., technicians who need to be able to log into a system and fix an error or patch an unexpected issue. To where one could build in controls to see somebody has to authorize it, etc. But it's really hard to do so because there's legitimate reasons for code to have vulnerabilities, for backdoors. AI is a transformative technology. We all agreed on that, from a national security to economics. We want the world running on that tech because the tech's important and because our values will be baked in. If you ask questions related to Tiananmen Square on Chinese models, you're going to get a very constrained answer that matters a great deal. So that was one set of goals. Another set of goals is to ensure that we dominate for this technology's use in the national security world. America had an edge, continues to have a small edge related to our chip design and production capability. That's really an allied capability, actually, the US, the Dutch, and Taiwan. And export controls was a policy to ensure those advanced chips didn't get into China's hands. I think everybody agreed that's valuable. The question you framed is exactly right. We don't want one or the other. Ideally, we want both, and how we form a policy that enables us to have both is really the most important goal. And then the final piece is, which, you know, I've led large organizations, and that's why one question I always think about when we're developing policy is, is this implementable? Because if a policy is not implementable, if it can't be enforced, then what's the point? We need to rethink the policy to perhaps make it less ambitious but fully enforceable. And I think that's always been a question in our export controls. I recall the Commerce Department, you know this better than me, Naveen, because you were there. I recall the Commerce Department saying, when you have so many different layers of export controls and so much of it, and we still haven't introduced, you know, technology to help us fully do this mission, it's going to be really hard to enforce it. So let's start with the question of what do we most want to enforce, optimize for that, and then use that flywheel to ensure our policy both achieves both goals of America, Western tech being the foundation the world builds on, holding our adversaries back as much as possible, recognizing they're also going to keep working at it, and being able to enforce it in a way that really ensures us what we're trying to accomplish is happening on the ground.

So to your point, you know, China's industrial approach incentivizes a lot of companies to produce, and the best wins. So you saw that in auto manufacturing in China, where massive number of brands all working to build the best cars, the cheapest cars, and get them on the road as the Chinese economy modernized and people started buying their first and second cars. Now, that institutes a winner-take-all, really aggressive domestic market in a way, to your point, that people don't often think about when we talk about Chinese subsidies. That is one of the drivers for one of the challenges the international trade economy is dealing with right now, which is China's surplus production is driving it to export and export cheap goods, right? Often for legitimate reasons that I just talked about, that are undercutting production around the world. President Trump's tariffs close the domestic market here, which is why Europe has been dealing with a massive amount of Chinese exports to Europe that are now really challenging Germany's auto industry, Germany's, you know, middle-class manufacturing industry of precision parts and engineering. And in the robotics case, Chinese advancements now to where they could build really precise actuators is starting to undercut Japan's strength in robotics, which is why there's a key interesting area for US-Japan partnership in robotics. But that's one part of China. The second part, which to your point I wrote about a bit both in the robotics piece and in the quantum piece, right, is this really cool concept of superclusters, ecosystems of software, hardware, in some cases academia, that come together that enable R&D ideas to make their way from the lab into manufacturing, into scale, into production. And that flywheel, we see it in robotics in terms of the rapid advancement of robots into Chinese manufacturing. We see it in quantum, in Hefei, in terms of the university system going to some of the Chinese government tests in quantum key distribution. You know, in terms of going to some of the progress that's happening there in quantum sensing as well. And I think that's a really interesting idea for us. Because in some areas in this country, we already have core clusters in Colorado, in Maryland, in California, building upon those, and some of them popped up around university ecosystems. But in having the incentives so that companies can quickly start there, iterate in terms of deployment, would allow us to take our strengths in capital markets and innovation and really use that Chinese model to compete better.

I think to your point, Naveen, sometimes they happen naturally, but often a bit of purposeful incentives helps ensure that it happens in the case that the natural knowledge doesn't. So quantum is a good example to where, you know, it crosses multiple technologies: physics, computer science, statistics, material science. And those are areas where we already have strengths, right? And then there are some universities that have quantum programs. But what you saw in Chicago, where there was a purposeful partnership, and actually former mayor Rahm Emanuel, when he became the ambassador to Japan, facilitated quite a bit of this. Japan has always taken quantum seriously, right? You know, there's a partnership between IBM and Japan in terms of building out a quantum computer. And what Rahm did was say, let's be a bit purposeful. Let's strike agreements between University of Chicago and the universities in Japan that already have programs, so that these are in some cases costly to build out the first R&D. Let's leverage each other's strengths. Okay, quantum algorithms require new ways of thinking about using quantum computers. Let's ensure those universities teach a class or two on how you code a quantum system. Some advancements in material sciences, quantum computers often have to be, you know, they're cryogenic-related capabilities. Where are those? Let's incentivize them to come to the same place. So I think some purposefulness, particularly when the initial technologies cost and require investment, help ensure that the cost of that investment capital knows one place to go and can leverage the various players who need that investment is useful. The question I have is, are there other enablers that cut across these technologies that we can develop a national consensus on? Manufacturing is such an interesting area. You know, you and I just spent some time with a number of founders, and we all bemoaned the loss of manufacturing in America in response to globalization, right? We really couldn't compete on the cost of labor, the scale-out of, you know, manufacturing and these superclusters where all the different parts of different kinds of components all the way through can be brought together in assembly lines. And we know that for America's both economic future and our supply chain security, we need some of that key manufacturing back. But manufacturing coming back is going to look different from what it was when it left 20 years ago, in the last, you know, in that period. And that's because today, you know, I had a conversation with a startup focusing on AI and additive manufacturing. And his point to me was, and I can train, we have a shortage of skilled labor in the country and the kind of manufacturing they're doing. He said, using AI, I can scale skilling far more quickly because a lot of the repetitive tasks AI machining can do, and then we can help the human in that human-machine teaming. So, and using robotics, programming robotics, programming robots so that they can do the parts of the work that frankly are often difficult for a human to do. Heavy parts, repetitive parts. So, I think that's where the conversation about manufacturing coming back to America to some will probably be disappointing because it's not manufacturing bringing large number of jobs back. It's manufacturing that brings back what we need here, particularly in strategic supply chains. But in a way that enables us to compete on cost. Because frankly across China we're now seeing dark factories, completely automated factories, right? If we want to compete on manufacturing, we're competing on those costs, right? And it's leveraging technologies that allow us to adapt that manufacturing as it needs new things. As a part needs to be more agile, bear more or less weight, come up with a different new material that's more resilient.

I think you're exactly right, and I'll pull the thread on it in the area of defense technologies because we haven't talked about that. We've talked about manufacturing in a traditional sense, and in the area of defense tech, you know, it's funny. I came into government in 2007 as a White House fellow, and at the time there was the surge in Iraq, and Secretary Bob Gates had made a commitment. The primary killer of service members in Iraq was roadside bombs, and there was a technology, a V-shaped hull to a vehicle, MRAPs, that was really saving lives because it deflected significant parts of the blast from the people inside the vehicle. And he made a commitment to scale and get MRAPs out rapidly to deal with the roadside bomb problem. And he turned to a group of us and said, 'Figure out what parts of defense acquisition need to be waived because they actually add cost and time and they don't, you know, they're not testing it to make sure it's a more resilient tech. Are we optimizing for speed and good enough? Get them out. There's lives to save.' And I spent a good deal of time studying the acquisition process. And I recall like unfolding this massive chart, you know, on my desk in the Pentagon, looking at it and the many steps and having this moment where you looked at it and I said, 'My, it's not optimized for speed and cost, right? It's optimized for other things, you know, the precision of what's developed, perhaps manufacturing across as many states as possible and leveraging jobs. And that could be fine. It wasn't fine in wartime. And it wasn't fine at that moment in time.' And the reason that mattered is because when you look at any number of acquisition reform studies in the 20 years since then, they say the same thing. That hit us hard when after Russia invaded Ukraine and we started sending American platforms. And we saw pretty quickly their precision targeting declined rapidly when they actually were in a real-world Russian electronic warfare environment. Our secure communications weren't, you know, weren't quite as secure as we expected. So that taught us that the changes that happened in software actually needed to happen in defense tech. Now, a couple of years before, venture capital had come to the same conclusion and started to invest in defense tech, which completely flipped the way procurement had always worked. As you know so well, procurement had worked where the military comes up with a set of requirements, goes out, industry builds precisely to those requirements. It did not really incentivize innovation because you could have some small contracts, but essentially the requirements process was the next gen of what we already did. And when venture capital started investing in, often people who had served came out and said, 'This is a gap I see. Software lets me build it more cheaply, more quickly, at scale.' And when venture capital started investing in that, you saw suddenly technologies that were some degree of testing and R&D already coming into the military and starting to get operational testing. And I think the wars in Ukraine, the wars in Gaza significantly accelerated that. So it's an exciting time. We saw some of that on our trip as well. In autonomy. You know, when you look at an autonomous ship, it's costing, let's say, roughly, you know, $1 to $2 million a ship. Doesn't have people on it. So a lot of the safety and security and resilience buildout you do when there's humans on board are not there. They're ISR platforms in some cases, they're weapons platforms. You can scale whether it's a counter-narcotics mission on a long maritime border or whether it's competing with China's maritime militia. You can scale capability in a way that wasn't possible before in the traditional way the defense built.