The Next Era of Warfare is Here

Thanks for having me on.

So several of the co-founders, so Matt Grim, Trey Stevens, we had been longtime friends and had worked together for quite some time. And then Trey had left to go do investing, working at Founders Fund, looking for a defense company to invest in. And at the time, right, this was 2015, 2016, nobody was touching defense in Silicon Valley. And so we'd been talking for years about doing this idea of, you know, bringing the best in Silicon Valley technology to the Pentagon. And you know, it seemed like a, you know, something that was, you know, a massive opportunity, massive impact on the mission side, and just nobody was willing to touch it. And so when Palmer was out of Facebook, he was like, all right, let's go. And we'd gotten to know him over the years as well. And so it kind of, you know, was this very unique opportunity in terms of the team, the size, and the ambition of what we were trying to do. And the fact that just nobody would do it. It was just too controversial at the time was actually, I think, a huge advantage for us in terms of being able to work on this problem.

So I spent probably about a decade working around the national security space, and when you meet these folks working on these problems, they are true believers in what they are doing. The mission is very visceral to them, very real. And providing these, you know, soldiers, sailors, airmen with the right capabilities just felt like something that, you know, the idea of just not doing it and saying too bad, you know, like you just get what you get, it just didn't make sense to me. And I think my other view of this has been, you know, the goal of the company is deterrence, right? Like the goal of the company is to not get into a conflict. So, you know, why do people get into these wars, right? They think they're going to win, right? And the history of warfare is often these things are, you know, very whatever your political aims were at the beginning, it's often quite unsatisfying. You don't quite get where you wanted to go. And so military power, you know, the idea is you want to make it clear what the outcome of these conflicts will be. You want to make it clear that, you know, use of force probably will not succeed, and that we want to provide our allies and the United States with the best capability possible to make that very clear. So 2017, it was a very different time, right? We just had not had significant large-scale conflict in a very long time, and so it was sort of hard for people to get their head around that this, you know, this could be an issue. Fast forward, you know, nine years, it's a different world, right? We've got multiple hot wars in Europe, or we've got, you know, kind of Russia-Ukraine. We've got everything going on in the Gulf, which has been persistent and, you know, will be for a long time, for better or for worse. India and Pakistan, Cambodia and Thailand, lots of, you know, kind of aggression in the South China Sea. Like, it's a very different time. And I think the understanding that hard power is a backstop for all the other peace and stability that you want is, I think, becoming more clear to folks.

I think a lot of the meta lesson from the Silicon Valley from the last 20 years has been going faster is more important than thinking, right? You have relatively few decisions you can't iterate your way out of, and you know, just the speed of learning trumps any sort of quality of thought you had up front. That's just not true in certainly a hardware world and especially in a defense world. So in a hardware world, you know, you make these decisions around what you're going to build, how it's designed. We got to ship a product that survives for 10, 15 years, turns on and works. It's got to work in just incredibly difficult conditions. And the cost of iteration in a hardware world is quite high, right? Both in terms of actual dollars out the door, but just timeline, right? Like, you know, if you've got parts that take three to six months of lead time, you don't get to, you know, that many iterations before you want to ship. And so the quality of thought, the foresight you have to have around these things, the level of engineering and analysis is quite different, right? And even as you extend that down into the software side, when you start talking about safety-critical systems controlling an aircraft, it's a different type of coding. It's a different type of operating. And so we've tried to blend the cultures as effectively as we can. Now, one of the pieces is that being rigorous, being disciplined is not the same thing as going incredibly slow. Like, it's just you can still do this efficiently, right? Like Elon has proved this out in a number of different cases. You know, I think we've proved this out in a number of different cases is that you can do that engineering quite rapidly. You can do your homework. You can really assess, you know, are these systems going to work and have a rigorous, you know, kind of testing and evaluation process. But that doesn't have to be 10 years long, right? That can be, you know, kind of measured in a much shorter duration in months. And so, you know, we've tried to blend the two cultures as effectively as we can. And I think we've done quite a good job with it. But, you know, we kind of reject the idea that this has to be a slow and lethargic industry.

So we started out working on a lot of things around border security, and this is one of the first areas we worked on where when we started out, we believed that modern technology, particularly, you know, computer vision, sensor fusion, how you would network all these systems together, would give the ability to make what is effectively a long-range nest camera, right? Like it tells you, hey, here is a person walking or an animal or a car, and gives that awareness to border security, border patrol officers, so they know what is going on. And this has been used, you know, we have maybe two-thirds of the southern border covered now. It's been extensively deployed. It's working quite well. And it's just providing that awareness of what is actually happening. You know, we've had a huge amount of support on this, you know, over the years. And you know, from there we extended into counter-drone technology. So this was things like, you know, great, we've got this technology that can detect things on the ground. How do we upgrade the radar, upgrade the cameras, add in some additional sensors so we can, you know, very robustly identify and track any drone incursions that are happening. And then it got to how do we defeat those? So we worked on a variety of jamming systems and surface-to-air missiles. So we have, you know, both kind of low-end versions of this with our Anvil system, which deal with, you know, kind of quadcopter threats, through to higher-end versions that deal with, you know, some of the Iranian drones you're seeing today. And then we've really expanded, you know, from there it was a lot of autonomous platforms. So this was everything from smaller drones into, you know, we did a co-development program with the Royal Australian Navy, make a school-bus-size autonomous submarine. We won a competition to do a loyal wingman fighter jet. So this is something we, you know, beat out Lockheed, Northrop, and Boeing. So us and General Atomics are building the Air Force's next-generation autonomous loyal wingman. It's an incredibly cool project, going really, really well. That was kind of a turning point for us where then we're proven on these high-end programs. We could design and build aircraft. We could do the autonomy software to power it. You know, we've built up a number of sensors over time, and since then we've seen a huge expansion in what we're doing, everything from space through to, you know, a very large portfolio of missiles and a variety of weapons, which, you know, I think was timed with the realization of some of the production challenges that are in the existing industrial base. We really just kind of expanded out year over year on the sets of problems we're working on. And I think, you know, there's a couple cross-cutting things. So one is from that software layer, we've really thought about what does the future of this look like? You are going to have thousands of systems on the battlefield, be it sensors, surveillance drones, strike munitions, soldiers, you name it. The battlefield in the future is much more distributed, much more complex, with a huge amount of information flowing in. So we built our Lattice software to network everything together, to fuse all that information, to apply AI and machine learning at the edge, and that's worked incredibly well. So now when we start a new program, you can bring to bear a huge suite of sophisticated technologies that make these systems very, very smart. And I think the second part of this is we've really focused by necessity on just execution and delivery. You know, I think one of the most scarce things really in any industry, but especially in defense, is you just do what you say you're going to do. Like actually just delivering and doing what you said on time, on budget. That's quite rare, and we have a track record of standing behind that. When we sign up for something, we make it work, we make it go. And I think that reputation has served us well. And our pattern of delivering very well has really what has accelerated us over the years. And now we're moving into a phase where we've won all these programs, we've built them all, built all these systems. We're really seeing, you know, massive ramp in production and manufacturing scale. And that's the sort of current era we're entering into.

Well, I think the question is how do you stabilize the world? And, you know, from our perspective, you know, you have a lot of different trends occurring. You know, one, there's a technology change, right? So you've got the ability for nations, you know, historically making like a cruise missile was something or ballistic missile that was something very few nation states can do. The degree to which that has proliferated is massive, right? You see what Iran has done, the number of other nations who have access to this technology, it's, you know, highly destabilizing. You've got the ability to do ubiquitous space-based surveillance at a relatively low price point, right? And so nations are able to understand everything that's going on. And, you know, obviously you've seen with Ukraine the rise in drone warfare and sort of these, you know, smart systems makes it a very, very dangerous time, right? So the technology shift has radically shifted warfare. Adversaries are going to see that as an advantage while our allies have not yet caught up, don't know how to defend it. And then the other side of this is you have a shift in US posture where for a very long time the US was so dominant, so clearly ahead on military technology that the idea that you could even try something, you know, combined with a fairly interventionist US policy meant that if anyone started conflicts, the US would come in, stabilize the system and say, 'No, we are not doing this.' Right? And so there was sort of this Pax Americana period was a very long time. We relied on it in all sorts of subtle ways that you're starting to see erode, right? Like I think people forget that before World War II, you couldn't ship anything on the high seas. It was just piracy. It was madness. The US Navy and the Allied navies stabilized the system to give us this period of global trade and prosperity that we've all been benefiting from. So the question with this is not, you know, sort of, you know, are we profiting off the war side. It is about stabilizing the world order so that our allies have the right capabilities to deter a conflict. Right? So we just did a weapons deal with Poland to provide them with strike capabilities. They're not going to shoot first, right? Like they're not going to antagonize Russia, but them being unprepared to counter a Russian attack, that seems like it would encourage Putin to attack more. And so you just got to make the consequence of these, you know, aggressive actions very, very clear. And our belief is everyone in the world is wrestling with this problem right now. How do they stabilize? How do they make clear to their neighbors that these attacks won't work? How do they build the right defensive capabilities as well as the right, you know, threat of offensive capabilities that will cause everyone to not attempt belligerent action, you know, thinking it might work.

I do agree when you just look at it from any dimension. The number of regional conflicts, you know, the number of simultaneous threats going on, this is very different, right? And it is, you have to look all the way back to around World War I period to find this rate of regional conflicts. So I do think it is a very precarious period in time. You know, the reasons for this are sort of hard to say. My best thesis on this is I think it is the combination of the technology shift with the political shift, but I suspect it is probably more skewed towards the political shift, which is, you know, Russia both expected that they would, I think they underestimated the technology change when they attacked Ukraine, right? Like everyone, New York Times, everyone believed that Russia was going to roll through Kyiv in three days, right? Like it was, that was the consensus. So I think they underestimated the technology shift here on the sort of defense side. But the political side, I think they believed that the West would not have the appetite to intervene. And I think this was a combination of US posture, where kind of Europe and NATO had gotten very comfortable, very believing that these conflicts weren't going to occur, had underinvested in defense for a long time, and had truly become quite reliant on the US in this regard. And so I think, you know, most nations, because they were able to rely on the US for so long, had atrophied their defense apparatus. And I do think that creates the conditions where, you know, a more aggressive, illiberal neighbor can take strategies to undermine their neighbors where they want to for their political ends, and the consequences don't seem that severe. And so I think that's probably the biggest shift that has occurred, which is, you know, the belief that they can succeed at these things or that through, you know, military or, you know, covert action they can accomplish. They can undermine the concept of NATO. They can tear lines into, you know, tear seams into, you know, kind of the allied construct. You know, they've been doing this in the information space with no consequence, right? By the way, no consequence for a very long time. And now are pushing the limits of what is possible on the military side.

there's a lot of different dimensions to this and there's a lot of ways to group it. You know, you've kind of got when you look at Ukraine and Russia, the small drones were largely changing what the, you know, kind of assault counter-assault. So like sort of for that military infantry, you know, kind of army infantry sort of scenario, that's a world that not many people want to even imagine right now, right? And so I think the reality is it's quite clear that, you know, with kind of short-range drones, the efficacy of sort of a military invasion, like an army invasion, is quite low, right? Like for a well-defended adversary and the cost of building that defense isn't that high. So I think that's like dramatically changed the equation where if like, you know, historically it was sort of the rule of thumb was three to one, you know, on the attack side to the defender side in terms of manpower and equipment, you probably need like a hundred to one, right? And it's sort of just so costly and so obviously costly that was a strategy that maybe sort of closed for quite a while. So then, you know, what we've been seeing is a lot more of what, you know, I would sort of call like long-range strike. So this is the ability to use missiles to inflict a level of, you know, destruction on your adversary. And what are they targeting? A lot of the theory before was sort of attrition of military equipment, right? So I'll win a military conflict by taking out my adversary's military. That's not what's happening in Iran. Iran is targeting the economic infrastructure. They closed the Strait of Hormuz most visibly. They struck data centers. They struck oil and gas refineries. And they understood that creating that economic pressure, that is what's going to cause the most, you know, sort of political capitulation on, you know, certainly the Western side. But you see Ukraine doing this to Russia as well, where Ukraine started doing long-range strike against oil and gas, you know, reserves and refineries. They're taking out the airfields and the supply chain in the rear, and they're striking all the way into Moscow to make it a much more challenging political problem for Putin to argue that this is working and that, you know, the defense is working and is this really worth it. And so this, the sort of category of long-range strike is, I think, the area that we'll see over the next several decades is the preponderance of, you know, sort of where militaries kind of turn on this for thinking about, you know, how to, you know, how to force that political outcome. And so that drives a lot of investment into long-range weapons and then into the defenses of those as well. And so there's a whole lot of consequences that play out from there. But I think the fundamental challenge, you know, that I think we're hearing from a lot of nations we're talking with is now you have a whole-of-country defense problem where your adversary gets to choose when they're shooting, you know, 100 rounds at one facility and you have to defend everywhere in your country simultaneously against those 100 rounds and you won't know where they're going to come and when they're going to come. That is a very hard math problem as the defender. And so I think this is the biggest risk on this. Then you have all these other sort of, you know, kind of spiderweb-style attacks where you covertly sneak in large amounts of drones to do, you know, kind of a different form of this strike against, you know, kind of infrastructure in the rear. But this is where I think a lot of warfare trends, which is really looking at if war is meant to effectuate a political end, this sort of approach seems to be the most effective at changing the calculus.

My view is these are very complex, nuanced policy questions and that writing two sentences into a contract does not actually lead to a disciplined policy debate weighing out all the competing objectives. Nor was there even an understanding of the wide body of policy that already exists within the US government around control of these technologies. So take something like the lethal autonomous weapons. We've had lethal autonomous weapons since probably the 1970s. You know, you look at the battleships we have, there's this Aegis combat system on there and that system has been able to go in full automatic mode for a very long time. You have an inbound threat and you don't have a lot of time to respond, you can have the ship do the work of determining is this a hostile or a civilian or a friendly military aircraft and then appropriately taking the response on that. You know, there's obviously risk there, but people have been more comfortable on the defensive side for a lot of good reasons because time is of the essence and, you know, you are by and large being attacked. So that, you know, that is one category we've had and that technology's gotten better over time. We've had this with torpedoes or anti-radiation missiles. So an anti-radiation missile you send out and you say here's a geographic area and if something is emitting like a Russian radar then that is a legitimate military target within that area. And so we've had this notion of, you know, you can delegate kind of the limited decision-making to machines in a bounded way for a very long time and I think that's realistically the framework of how this has to work. But importantly with that there's a human agency that exists, you know, so the US government has always had this policy which is that humans are accountable for the use of force in a battlefield. You send an anti-radiation missile out and it picks the wrong thing. That's on the command, right? That is a, you should have known what that weapon was going to do, how it's selected, and, you know, were you safe to employ that weapon in that context. That has been the underlying framework that the US and our allies have had, which is this is fundamentally about human accountability at the end of the day. You want your weapons to know exactly what they are going to do, how they are going to make decisions, and what bounds you put on them. So the debate really boils down to, you know, should you use AI to get more precise, more accurate, more discriminating, or should you go with older generation technologies that are, you know, dumber? Right? To me the ethical imperative is to make the smartest, most discriminating weapons you can and do it in a way that is responsible based on where the technology is. And that's everything I've heard from the US government as well. They are not interested in a, you know, fully automatic AI runs World War III Skynet scenario. That's not even on the table. What they are interested in is how do we make these systems more precise, more targeted, more discriminating. And the US has led the way for decades in looking at this question of how do you employ weapons in the most accurate and precise way with the least collateral damage. And that is something they have spent tens of hundreds of billions of dollars investing into. When, you know, our adversaries like Russia are just happy to use dumb bombs, right? And that is a big difference. So to me the imperative is really how do we provide the best technology to make these things as, you know, precise and limited as we can. The alternative seems just wildly crass that you believe like they should have dumber things that are less safe to use.

Maybe starting with the first. I think Anthropic actually does work with the US government extensively. They were the first ones leaning in. They have extensive work with the intelligence community, with the department, well before this came out. I think their concerns on, you know, use of these technologies in a way that the technology is not quite ready to be responsibly used is legitimate. But then I think the way you engage in that is actually working through those policy questions, being highly involved and steering the government away from things where they're not technologists,

Right? Like they're policymakers, they're war fighters. They need technologists to tell them what can and cannot be done and how can the technology get there in a way that is responsible and reinforce the values and norms we have. And so I think in reality, they were working with the US government, continue to do a lot of work with us and several others, and you know, I think they are actually, you know, kind of very mindful of providing the US and allies with the best technologies possible. So, you know, I think what went wrong was probably more a matter of personalities, trust, and, you know, a style of engagement to get to the outcome on this. And, you know, I get these things are hard. They're very complicated topics, and it's very tricky to work through. You know, maybe on the second question, you know, the US actually has put out several different detailed things on what went wrong in that, you know, mistakenly attacking a school where they had multiple different databases. Some of them had marked it as a school. There was some older ones where the IRGC had used that as a command and control center in the past. And so, you know, this is where I think using technologies like AI to be able to coalesce all that information, to be able to have the most up-to-date information, are massively impactful. Where when you think about the sheer volume of things that the US was doing leading into that strike, where they did something like 10 times the number of strikes that they did in the Gulf War within the first month, they executed like a thousand of these in a day. And for each one of those, they have a process by which they validate that the intelligence indicates that this is a target and that there is, you know, collateral damage assessment and, you know, all these, you know, kind of legal factors. One of the only militaries in the world, by the way, that does that or cares to do that, right? I don't think Iran is running that calculus every time they take a shot. And, you know, they have gone through and traced through what went wrong and what's happening. Now, I'm not going to sit here and say like, you know, how Trump presents these things is, you know, the most measured way. Of course it's not. It's President Trump, right? Like that is the reality. But the US military as an institution has gone through and traced through why this happened, what went wrong, and how do they fix this in the future. That is a sense of accountability on this that I think is what we're looking for. There will be mistakes. War is terrible. Again, this is why you do not want to get into war. You want to create the deterrence in advance. And the reality is these things are, you know, having that discipline to actually improve how you do this and investigate what went wrong, that is something that I think the US leads the way on.