AI warfare and the race for a million-drone army

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

TSENG: As a former Navy SEAL who has had to make the moral decision about the use of lethal force on the battlefield, I fundamentally believe that decision is 100% a human decision. Every modern military is going to proclaim a million-drone army, air force, navy, and Marine Corps powered by AI and autonomy within the next five years. Today, a team of two engineers using our developer tools can build an AI pilot for a one-way attack drone in less than two weeks. Four years ago, that would have taken 150 engineers 18 months.

SAFIAN: That’s Brandon Tseng, co-founder and president of Shield AI. Shield is a leading maker of autonomous military technology. Its drones and software are used from Ukraine to the Middle East. I wanted to talk with Brandon about the boom underway for defense tech firms and the rising debate over AI in warfare.

Brandon is a former U.S. Navy SEAL who deployed in Afghanistan, and he shares both his optimism and areas of concern when it comes to autonomy on the battlefield. Wherever you stand on the use of AI in modern warfare, Brandon gives us a timely glimpse into a high-stakes reality with escalating impact. So let’s get to it. I’m Bob Safian, and this is Rapid Response.

[THEME MUSIC]

I’m here with Brandon Tseng, president of Shield AI. Brandon, thanks so much for being here.

TSENG: Thanks, Bob.

Copy LinkWhy defense tech went from fringe idea to strategic priority

SAFIAN: You started Shield AI with your brother Ryan a little over a decade ago after serving as a Navy SEAL officer. You deployed twice in Afghanistan and saw firsthand how robots could gather intel more safely. For a while, though, defense tech was kind of shunned, particularly in Silicon Valley, because of concerns about autonomous weapons. That’s changed. Shield AI is valued at almost $13 billion. Do you feel vindicated, or do you still feel underappreciated?

TSENG: You know what’s funny? I would actually say defense tech, quote-unquote, didn’t exist in 2015. People couldn’t even shun it because no one was talking about it. I’ll tell you a quick story. Peter Levine from Andreessen Horowitz sits on our board. He’s an incredible investor and also a Stanford professor. He gave a guest lecture when I was at Harvard Business School. He said, Yeah, I just invest in dumb companies, and I just invested in the dumbest idea yet. And that dumbest idea was Shield AI.

I called Peter up and said, I heard you just invest in dumb ideas. He said, Yes, I did say I just invest in dumb ideas. It’s a dumb idea to get into a car with a stranger. That’s Uber. It’s a dumb idea to stay at a stranger’s house. That’s Airbnb. Everybody knows it’s a really dumb idea to work with the government to build defense. He said, Now two things are going to happen. Either one, you guys are going to fail, and it will prove to be a dumb idea, or you guys will be successful, and everybody will say, “Oh, that was a really clever idea.”

So 11 years into the journey, I feel like I can breathe a little and say we were a little ahead of our time in terms of thinking about defense and the role of AI and autonomy in defense. So yeah, maybe a little vindicated. But there’s still a long way to go.

SAFIAN: Your autonomous military tech is being used in Ukraine against the Russians. It’s been used in Gaza by the Israelis. The core product is software called Hivemind that allows drones and other vehicles to operate without a human in the loop and without GPS. Was that the dumb idea? Was this the capability and impact you were aiming for?

TSENG: The original vision, and my background is I’m an engineer who has always been fascinated by technology, was shaped by reading a lot about AI and autonomy in 2013, 2014, and 2015. I came to the realization that the world is going to be full of autonomous systems. Self-driving cars, humanoid robots, self-driving airplanes like what Shield AI does, are really just the tip of the iceberg.

I would claim that within this century, you will see autonomous systems outnumber human beings on the planet. The impacts of that are going to be incredible. I think there’s going to be massive positive impact, and I wanted to be part of that. Then it became, OK, what problems can I solve? I kept coming back to the problems that I had faced and decided that every single military asset by 2035 should be powered, commanded, and maneuvered by artificial intelligence. That was the original vision for the company. And look, probably not every military system by 2035. Maybe it takes until 2040 or 2045. But at this stage of the game, the momentum is unstoppable.

Copy LinkWhat people get wrong about AI on the battlefield

SAFIAN: There is tremendous excitement around AI. There is also distrust around AI these days. Shield AI is in perhaps the highest-stakes arena, as evidenced by the controversy between Anthropic and the Department of War, Department of Defense. What do people most misunderstand about the role of AI in modern warfare?

TSENG: A lot of people immediately go to the sci-fi dystopian movies: The Matrix, Terminator, et cetera. And that’s not what it’s about. In the AI space that we operate in, it is very much about high reliability and high assurance. When we flew F-16s completely autonomously with our AI pilot, the U.S. Air Force wasn’t just like, Hey, go try this out and see what happens. You have to prove that you can do that with high assurance. You have to be able to certify that AI. You have to be able to certify that autonomy.

That’s what I don’t think people realize. This is not something that is going to escape out of the bottle, genie-in-the-bottle type stuff. It’s a well-engineered system with levels of assurance that it does what we want it to do.

Copy LinkHow cheaper autonomous drones are replacing legacy aircraft

SAFIAN: Ukraine is the theater where autonomous weapons have been the most prevalent, the most used. Your V-BAT drone is on the ground and in the air. How is the V-BAT different from other drones, and are there things you’ve learned from real-world deployment that are different from what you expected?

TSENG: V-BAT is powered by our AI pilot, Hivemind. The easiest way to think about an AI pilot is self-driving technology for unmanned systems. It allows them to operate while GPS and communications are jammed, which is every single mission in Ukraine. It allows them to autonomously execute missions, and it allows them to swarm or work together.

The easiest way to think about the V-BAT is that it does the mission of a Predator drone or a Reaper drone for a fraction of the cost. Predator drones and Reaper drones cost about $40 million. The United States has had 40 of them shot down since October 2023. We had 26 of them shot down by the Houthis — a $1 million weapon system blowing up a $40 million aircraft. In the conflict against Iran, we’ve had 14 of them shot down, again by really cheap weapon systems comparatively.

I used Predators and Reapers as a SEAL in Afghanistan. Fantastic weapon systems for the fight in the war on terror, but they are not assets that can be used on the modern battlefield with any expectation of success. And when they get shot down, you’re losing a $40 million asset, and you cannot quickly produce another one to get it out the door, even if you had the money.

V-BAT does that same mission for about a million dollars, a fraction of the cost. That’s really what you’re seeing in the defense industry and the opportunity for these defense tech companies. You have these legacy assets, and now this new generation of smaller, more affordable assets can do the exact same mission. It reminds me a lot of the compute industry, where my cell phone can do everything my laptop and my desktop can do. Desktops still exist in the world, but they’re far fewer in number because people can use their mobile phones. It’s not that different.

SAFIAN: A lot of people’s understanding of smaller drones is kind of limited to toys if they don’t have military experience.

TSENG: Predators are operated out of large operations centers. You’ll have colonels and generals watching the feeds, and they’ll be making decisions around the use of kinetic effects or lethal effects coming off those systems. Quite literally, our V-BAT is getting the exact same feed. I’ve actually watched operations from my desk here in Washington, D.C., and I was like, man, this is exactly what it was like for me in Afghanistan in an operations center doing the exact same thing.

Our V-BAT is vertical takeoff, launch, and land, so a Predator drone requires a runway. If you asked Henry Ford what the customer wanted, they would say faster horses. For a long time, the military was quite literally asking for faster horses instead of saying, What’s the best way to solve the problem? Our V-BAT takes off vertically and flies for 13 hours. You can be much closer to the problem, while still being far enough away to stay safe. You can get a lot more data because you can run a lot more operations for the same price. That’s how we think about it at the end of the day.

SAFIAN: This idea of ubiquitous autonomous drone warfare making us safer — I have to say, for some folks, the vision of millions of drones flying around overhead can sound a bit terrifying.

TSENG: I believe that AI and autonomy will be the most strategic conventional deterrent technology for the next 100 years. We have nuclear deterrence, and our nuclear triad deters other countries from committing nuclear war. So then you ask, what are our conventional deterrence mechanisms today? Our premier conventional deterrence mechanism is our carrier strike group, where we park an aircraft carrier off the coast of Iran or off the coast of China whenever we want to remind an adversary, You better think twice about invading a country.

Our adversaries have invested in countering the carrier strike group. They have invested in long-range anti-ship missiles that just go farther than our current deterrence mechanisms do. So in this new world, if you ask what this new deterrence is going to be, it is going to be millions of drones. Had Ukraine been able to put up even hundreds of thousands of drones on its border a week before Russia decided to invade, I promise you that Putin would not have invaded. They would have thought twice about the cost of invasion.

I would estimate that every country, every modern military, is going to proclaim a million-drone army, air force, navy, and Marine Corps within the next five years. If you can put up drones, it is a way of securing your sovereignty. It is a way of deterring your adversary. So that’s what’s happening.

Copy LinkWhy human judgment still matters in lethal decisions

SAFIAN: Leaving lethal decision-making to AI: Critics argue that removing humans from the kill chain is a moral red line. Different defense firms have different perspectives about that.

TSENG: Look, as a former Navy SEAL who has had to make the moral decision about the use of lethal force on the battlefield, I fundamentally believe that decision is 100% a human decision. That is Shield AI policy, that is U.S. Department of War policy, and that is NATO policy. Even our adversaries have claimed that they want that moral decision around lethal force to forever remain with human beings.

Then, yes, that system is going to help execute your intent — what we would call the commander’s intent — around that decision, and actually make the battlefield more precise and less destructive at the end of the day. Certain things can work like an on-off switch. I’ve got a Tesla full self-driving car. It’s never going to just decide to start flying. That’s not what it was engineered for. These systems will not just decide to make these decisions if humans have not allowed them to do so.

SAFIAN: What AI can do today is very different from when you started the company. How do you keep up? How do you stay ahead?

TSENG: It’s become lightning-fast. Today, a team of two engineers using our developer tools can build an AI pilot for a one-way attack drone and have it up and flying in less than two weeks. Four years ago, that would have taken 150 engineers 18 months. Now, Shield AI will say this isn’t just a lot of leverage from existing tools. We’ve had to build a lot of stuff ourselves. But the things we have built have made us faster than any company in the world at building AI pilots, getting them flying, and getting them on the battlefield.

We have a saying at Shield AI, and this came from our head of aircraft engineering, who was the lead engineer on Starlink, reporting directly to Elon Musk. He said, Look, the team is the product at the end of the day. I really do think you have to have an expert technical team, certainly at any technology company, and expertise in AI and how it will be implemented in your business.

Copy LinkHow fast AI development is reshaping military competition

SAFIAN: I want to pause here, in part to double-click on Brandon’s point that every company needs to make AI expertise part of its core strategy. But I also want to underscore the inordinate high stakes when it comes to defense tech and the intense pace of change. So how far along is China in its military use of AI, and what are the implications when AI defense tech is shared with other countries? We’ll talk about that and more after the break. Stay with us.

[AD BREAK]

Before the break, Shield AI’s Brandon Tseng talked about the rising use of autonomy and drones in military warfare. Now we’ll talk about China’s military advances with AI, trusting the U.S. military and its allies to use AI appropriately, and whether global unrest benefits the defense industry. Plus, why Brandon finds entrepreneurship as grueling as being a Navy SEAL. Let’s jump back in. I’ve had conversations with several CEOs recently who were talking about the prowess of China’s technology. Where is China on autonomous military tech, drone software, and otherwise?

TSENG: One, it’s really hard to tell. They have an incredibly opaque ecosystem, deliberately so. What we do know is that the U.S. continues to lead in AI, in autonomy, and in software. However, it does not matter if you are in first place. What matters is the rate of innovation. The challenging thing, and the thing that I would claim every American should be concerned about, is that China’s innovation engine is incredibly fast. At some point, they could surpass us.

This is supported by a number of different data points, including classified data points, that show they are iterating much faster than the U.S. But what I won’t discount is the American spirit and how competitive and hard-working America can be in getting after these problems. The average American might say, Why do I care if China has a better military than the United States? I don’t think people recognize or appreciate that American military dominance has been the prerequisite for the global world order that has enabled so much growth across the world for the past 75 years. The world looks very different if a country with a very different value system challenges that rules-based order because America no longer has that military dominance.

Copy LinkWhy defense companies reject the idea that war is good business

SAFIAN: One of the bugaboos about the U.S. defense industry always has been that people say it benefits from global unrest. Peace is bad for business, and war is good. The munitions deployed in Iran will be replenished, for instance, and that’s a windfall for someone. Do you wrestle at all with that aspect of being part of the defense business?

TSENG: No, I don’t wrestle with that aspect. It doesn’t bother me at all. I think it’s a misnomer to think unrest is good for these defense companies. That’s not why they do it. You have to recognize that we operate in the real world. And so long as countries have very different value systems, there is going to be friction between them. We all hope they can resolve that friction through diplomacy, but unfortunately, having served in war and seen war firsthand, I know war is often used as a way to resolve that friction or those disputes between countries.

It is a world where the defense industry can be very cat-and-mouse in the sense that, OK, these people have built a better airplane or a better missile, so we need to build a better airplane and a better missile. But it’s actually not that different from any technology company. It’s like, OK, they built a mobile phone, so we have to build a better mobile phone. They built a piece of software, so we have to build a better piece of software. It drives an innovation engine. We have a saying at Shield AI: The greatest victory requires no war. War is the most destructive force on the planet. I don’t want it. But I do want to help deter it and, if necessary, help our warfighters and our allies win.

SAFIAN: Unlike a lot of U.S. defense suppliers, the majority of your contracts, as I understand it, are abroad — not just Ukraine but Taiwan, India, and the Netherlands. Are there any complications in that dynamic, any pushback from the U.S. government about who you partner with, or things you have to navigate?

TSENG: I think it was about two-thirds of our revenue last year that was international. And the answer is no, we have not experienced any complications. The U.S. wants to see these countries using U.S. technology. There’s great benefit for a country like Taiwan, and for our European allies, in using the same systems the United States is using. There’s an interconnectivity that will inevitably happen if you require combined joint operations.

Copy LinkHow allies, exports, and trust shape the use of military AI

SAFIAN: I mentioned to a colleague that I was going to be talking to you, and they asked me whether or how much you monitor if your tech is being used for a, quote, “just mission.” I suppose you experienced that as a soldier too, in terms of how you were being deployed.

TSENG: The answer is, first, we work with our partners and allies. At the end of the day, they can choose not to share data. They can choose not to share what happens on mission, if they so choose. We obviously want to work with them. We want to get the data because then it can improve the product. They have a right to sanitize that data. But again, I’m going back to defense exports and why we work with these allies and partners. The State Department governs this. Not every single one of these countries has the same values, justice systems, or laws. But the reason we partner with them is because we say, hey, it’s better that they align with the USA. We can influence them over a period of time, versus having them align with China and China influencing them over time, or Iran or Russia. It’s much better to have them align with the United States. I have never come across a situation where I’ve said, “Hey, I totally disagree with the State Department.” If it came to a point where there was a big delta between what the State Department was trying to do and how I thought about the world, or Shield AI thought about the world, it’s something we could take a look at, but I just don’t see that happening.

SAFIAN: And you trust that the U.S. military’s use will always be just and appropriate. Because you’re in this business, it means that sometimes your services and your technology may be used in ways that maybe you personally might not have wanted, but you trust that.

TSENG: Yeah. Having been part of that machine, I absolutely trust it. And look, I’ve had to tell a couple of people this: The care that the U.S. military takes over the utilization of such incredibly powerful technology is astounding. For every kinetic strike, for every Predator Hellfire that we shot, we turned off the mission 25 times more because we said, “Hey, there’s a risk of collateral damage. There’s a risk of civilian casualties. We don’t know who’s inside that building. We’re not going to do this operation.” Every single one of those missions, you have intelligence analysts, you have geospatial analysts. They’re looking at the risk of collateral damage, the risk of civilian casualties, and the probabilities that the enemy is who we believe they are. I can’t think of an organization in the world like the U.S. military, where they have the weapons to dominate the world but also have extreme care and stewardship over those weapons and that technology. And look, they’re not perfect, just to be clear. The U.S. military has made mistakes. But I’d call it 1,000 times more wins than mistakes.

Copy LinkWhat entrepreneurship and Navy SEAL training teach about resilience

SAFIAN: I saw somewhere that you said entrepreneurship, day in and day out, is grueling. It’s awful.

TSENG: Yeah, it is awful.

SAFIAN: So is it better or worse than being a Navy SEAL? And are there things that businesses could learn from Navy SEALs, and things the SEALs could learn from businesses?

TSENG: The suffering is pretty equal. I would stress out more over losing investors’ money. That’s something I never wanted to do. I get more stressed out over that than the missions we would do as SEALs, and the risk to life as a Navy SEAL.

SAFIAN: Because it seemed less risky because you were prepared in a different kind of way than in entrepreneurship, where you don’t really know what’s going to happen?

TSENG: No. I think it was because I had all my teammates, and we had all gone through training. So I’m like, oh, all these people signed up. We know the risks. We all have the exact same information. And look, an investor has lots of information, but they don’t have the same information, the same experiences, or the same market knowledge. You try to explain and talk through that, but it was just different in that sense. Especially the early investors, and myself included — I didn’t know all the risks of going into the defense industry when I started.

SAFIAN: Are there things that you reflect on from your Navy SEAL days that you apply to the business?

TSENG: Oh, 100 percent. One is the cultural aspect. I tell people Shield AI has a warrior culture. I would define that as a highly professional, highly disciplined organization that pursues excellence, figures out how to get the mission accomplished, and doesn’t give up. Taking it one step at a time is so important. These are things that you learn in the SEAL teams. When you get knocked down in entrepreneurship, you get back up and move forward. So I think there are a lot of parallels between the experiences I had as a SEAL and what it takes to do what we’ve done at Shield AI.

Copy LinkWhat is really at stake in the global AI arms race

SAFIAN: What’s at stake right now for you — not just for Shield, but for all of us, for global defense right now, and where we go from here?

TSENG: What’s at stake is the global world order. Stability and peace are fought for and earned every single day. There are underlying technologies that will enable other countries to challenge the status quo. Again, I think it’s so important that America leads here. If China came out and said, “Hey, we’re building a 100 million-drone army powered by AI and autonomy,” it would terrify the U.S. And I don’t put it past them to make that announcement at some point in the future. So I think it’s so critical that the U.S. is prepared. Anytime there’s a new technology, militaries will adapt to it. The ones that do so successfully, their governments get to shape the world. And again, I think it’s been a good thing that the United States has helped shape the world over the past 70 years.

SAFIAN: Brandon, thank you so much for chatting about all this with me. I really appreciate it.

TSENG: Really enjoyed the conversation.

SAFIAN: Talking with Brandon, it’s clear that the term arms race isn’t just a metaphor. It is underway right now, in real terms, with AI at the core, all across the globe. And whatever your perspective on the defense industry, it’s a race with massive implications for our future. The need for a human in the loop with AI may seem obvious, but as Brandon underscores, it is critical. Autonomous military systems are a reality, and the choices we make with them really matter. So much comes down to trust. It will be on the government, the military, and businesses like Shield to earn that trust, not just every election cycle or new conflict, but every day. I’m Bob Safian. Thanks for listening.