The a16z Show - Emil Michael: Iran, Anthropic and the Future of AI at the Pentagon

Episode Date: March 13, 2026

This conversation with Emil Michael, undersecretary of defense for research and engineering and acting director of the Defense Innovation Unit, was recorded at the a16z American Dynamism Summit in Washington, D.C. Michael walks through how he inherited a department running 14 undefined technology priorities, cut them to six, and made applied AI number one. He also gives the first detailed account of why commercial AI contracts written under the previous administration created a vendor-lock crisis that put active military operations at risk.     Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

We're faced with the biggest military buildup in history. We're trending toward artificial general intelligence, a substrate, a layer, something that'll touch everything. But we're way behind an AI at the department. You are a CTO for the Department of War. How do you take stock of where those priorities are? When I took the role, we had 14 critical priority areas. We got them down to six, and there were the places where I thought we had the greatest opportunity for change and for growth. impact. There has been an incredibly public discussion about commercial AI models being used

in the Pentagon. What has changed in this latest discussion? I had a holy cow moment because there were things well beyond what you've been hearing in the press in the last couple of weeks. When Emil Michael was confirmed as Undersecretary of Defense for Research and Engineering, he took inventory. What he found was a department with 14 critical technology priorities, most unchanged for nearly a decade, written in a language so vague, no one could act on them. He cut the list to six. Applied AI went to the top. Within 90 days, 1.2 million of the department's 3 million personnel had used some form of AI. When he started, that number was 80,000. The more urgent problem was what he found inside existing contracts. AI models baked into

the most sensitive commands in the U.S. military, under terms that could shut the software off, mid-operation. A company's internal values document, he argues, cannot be the governing authority for American command and control. This conversation with Emil Michael, under Secretary of Defense for Research and Engineering, and Acting Director of the Defense Innovation Unit,

was recorded at the A16Z American Dynamism Summit in Washington, D.C. All right, thank you for being with us. I know every week is very busy for you, but it feels like this past week has probably been the most publicly busy for you. So thanks for joining us. My pleasure. Good to be here.

Look, we're going to talk about anthropic, AI, and defense, but I think we want to talk first a little bit about you, how you got into this seat. This is not your first tour of duty in government. You have decided to be a public servant before and for a long time. For half this room that comes from the technology side, they know you as an incredibly accomplished Silicon Valley Executive, highly sought after, very successful.

Let's start with what pulled you in? into public service. How did it start? Why do it and why take on this role now? You know, after my first company, Tell Me Network, so we had this speech recognition software we sold to Microsoft. And it was seven.

I kind of needed a break from tech. So I applied to this White House Fellowship program, which was super cool. Colin Powell had done it, chairman of joint chiefs of staff. General Kane had done it. It was just a cool program. It was a year, nonpartisan program.

And I got selected, which was awesome. And I got assigned to Robert Gates, who was the Secretary of Defense of the time. So I got to spend time in Afghanistan, Iraq, Pakistan. I got a passion for that world and it was so new to me. And I said, someday I'll come back when I'm ready to really go. And then President Trump got elected. And I was like, wow, this is a moment where you have a disruptor at the top.

He chose Secretary of Hexas. And that meant all the problems I'd seen there would be a clearer way to solve them. And you could see by the pace we're moving at, the executive orders, critical minerals, new tech. New entrance, the pace is incredible relative to what it was before. So I'm excited to be there. Awesome. Well, why don't you talk to us a little bit about that? You have been outspoken that the Department of War cannot operate at peacetime speed. This is kind of like how we tell founders. You can be a wartime CEO or a peacetime CEO. What does peacetime speed mean? What is the problem with it? And what does winning look like? What does moving with urgency look like?

Peace time speed started after the Cold War was over at the Department of World. war and I guess there was some famous event called The Last Supper where the leaders at the Pentagon said to industry, hey, there's not going to be a lot more buying of weapon systems or innovation. So you all should consolidate and slow down your growth basically and become dividend payers and stock buybackers. And that lasted for a long time. And then the industry consolidated down to four or five primes. So then we're faced with the biggest military buildup in history in China starting in the mid-2000s, like 2010, and we didn't catch up. So all of a sudden, we've outsourced a lot of our key domestic production on many different

areas, whether it's critical minerals, batteries, a lot of the supply chain stuff. And then you look up and you're like, holy cow, we've got a lot of catching up to do. So wartime speed is ensuring that redomesticating the critical, things we need for national strength. And we are moving out on that. We have exquisite capabilities. We have a lead in so many areas, but we have to make sure that we're self-reliant in key areas. And so we're running as hard as we can on that. You know, it was amazing to me. We work with a company called Skydeo, and they were sanctioned by China. And we usually hear stories about how companies are sanctioned by the United States from doing business here. But actually, if you're a company that's an

important company, and you get sanctioned by China and you can no longer buy motors or batteries or other really key things that primarily come from China, it stinks. And if those products make it into the defense procurement cycle and we can't get them anymore, it's a huge problem. So we've learned in our practice that it's not just the defense industrial base, but also all those precursors. So talk to us a little bit about as you stepped into the role, you are a CTO for the Department of War. How do you prioritize? How do you take stock of where those priorities are? How do you figure out? I mean, if you read the news, it feels like there's an endless list of things that we need to catch up on.

When I took the role in May, it was sworn in. But did the inventory like any new leader would do? And we had 14 critical priority areas in this role. And my predecessors had 10, and then someone added four. So they were 14, and they hadn't changed in nearly a decade. So I get to look at them and say, well, okay, well, why are these critical? And by the way, who can remember 14 things? and you're trying to motivate a workforce.

And what is an integrated network systems of systems? They were sort of techno babble. So I said, well, let's study and say what's really important. And we got them down to six. And there were the places where I thought we had the greatest opportunity for change and for growth and impact on our combat power and for our industrial base. And that was a starting point. And applied AI was number one.

So then we moved the chief digital and AI office into my group and we were able to run extremely fast with that because we're way behind an AI at the department. And we know it's penetrating the rest of the world so dramatically, so quickly. And our adversaries, we're also using it for lots of different purposes because they have less trust than their command and control. So they want to use computers to sort of eliminate human decision making. we want to enhance human decision making. So that's been a priority, and we've kind of, in 90 days, I think we've had 1.2 million

of the 3 million people at the department used some form of AI. That number was 80,000 before I started. Incredible. Let's talk a little bit more about why it's important to have advanced AI capabilities. Maybe some people would say, well, we've been really good at, you know,

being a world power without all this AI capability. Why is it important? Can't we just focus? focus on like making missiles and stuff? Well, I can help make missiles if you can help you solve physics problems, material science problems, aerodynamics problems, or opportunities. But we split up the AI sort of efforts in the three areas, like enterprise corporate use cases like any big organization would do for efficiency purposes.

And that's just normal, like goodness. Everyone's happier when you could do mundane tasks faster. And then intelligence purposes, which is we collect. enormous amount of intelligence. We have huge repositories of data at the department that's sitting there siloed unused. Imagine decades of satellite imagery that you could use to train a model and get insights and then use that to do anomaly detection so you could find out what's happening. So you take a human analyst and you increase their throughput by a thousand. And then for war fighting, war fighting they talk about as big part of it as logistics. How do you plan logistics and find assets and do war gaming and play operations and do simulations and so on. So there's a lot of really important use cases that we could benefit from. Yeah, I think we are the beneficiaries of startups that pitch us their ideas all the time. And because it's American Dynamism, we hear about companies that say, look, we're going to save 15% of the Pentagon's fuel budget through logistics by using AI. We're going to figure out

how do you move the troops and the equipment in the vehicles in the most efficient way in a contested environment. These are all amazing, amazing, incredible use cases. So look, In the last week, there has been an incredibly public discussion about commercial AI models being used in the Pentagon, including Anthropic, being used inside the department. In the old software days, and I'm going to really be in the old software days before even SaaS, people would just buy software and then the customer would use it however they want.

What has changed in this latest discussion? What has come up? What are the issues? I assume most people in this room have been following along, so I think it's safe to assume there's a baseline understanding. but what have become the key issues? And why did they come up now? Well, they came up now because as I started to look at the contracts

that had been written during the last administration for the use of AI, I had a holy cow moment because there were things not well beyond what you've been hearing in the press in the last couple weeks, things like you couldn't move a satellite. You couldn't plan, you know, an operation, couldn't plan it, not use it. to execute it, if it would potentially lead to a kinetic strike or something. Dozens of restrictions. And yet, these AI sort of models were baked into some of the most sensitive and important places in the U.S. military where we do exercise combat power.

Central Command now that's sort of era of responsibilities Iran or, you know, Indo-Pa-com, who's era of responsibility is China, or Southcom, which is the, you know, Venezuela and South America, we're all using this model. And there was no two vendors. It was a vendor-locked situation with terms that, in theory, if the model was designed to turn off when you violated the terms, could just stop in the middle of an operation and put lives at risk. So that was the one moment.

I was like, okay, we have to fix this and clean this up while we're deploying this in the department. So that raised all these issues. And then second, after the Maduro raid, one of the primary vendor for this had raised the question, senior exec, about whether their software was used during the Maduro raid, which is one of the most successful military operations of our lifetime. Like truly an incredible operation. And that sort of... Yeah. It's been a bad a few months for the enemies of America. It has been.

And I got to meet the guys who did that, the lead helicopter. guy who got the Malavana. It was a truly outstanding American. I mean, he was shot and kept his cool, didn't tell anyone shot so that the first landing team could land. And it didn't freak it out. And it could have blown whole operations of the courage was incredible. And when a company says to you, hey, was our software used there?

Because we're not sure we'd like that. Chill goes up your spine. It's like you're at a coffee shop and some strangers like, hey, I saw your kid at school yesterday playing cakeball. like, who are you? You know, what? So that set off a series of events of, whoa, are we single-threaded on a vendor who's concerned about how we're using the software after the most successful military raid

and the terms of service do not comport with the future world? We've got to get other partners in here and we've got to move. And if you think about AI as like what we're trending toward AGI, artificial general intelligence, a substrate, a layer. something that'll touch everything, like the internet touch everything, or the telecommunications network touched everything. Then to tell the users of that substrate of technology, you can't use it for legal things,

things that have come through the democratic system, laws passed by Congress, executed by the executive branch. In the military, which is the most sensitive part of the U.S. government, because our job is to be the strength department to protect Americans, you do have a moment of truth there, like we've had, which is this technology, if we're losing it lawfully, the substrate has to be our choice. It can't, the software, someone's soul of their model,

their constitution, which is not the U.S. constitution, can't be dictating our command and control environment and telling generals and we're fighters what to do and not do. That's right. So first of all, I think you just shared a bunch of backstory and information that, you know, has not been readily found, I think, by reading the news and following along, although probably a lot of people in this room could have guessed, but it's nice to hear that part of the impetus for this and the catalyst for this discussion was originally about making sure that you have ways to use this capability that fit in with our democratic norms. And, you know, you mentioned that, you know, we have a constitution and they have a constitution.

It's kind of crazy that a company even has a constitution. There's not that many companies, I think, that have their own... Corporate values, yes, constitution, no. Corporate culture, corporate values, for sure. I don't know how many have a term that they say this is our constitution, especially when there's often a revolving door of executives at companies. Like, you know, who knows what they'll think tomorrow. So talk a little bit about the role of democratic oversight,

elected leadership that sort of guides the process that we have in this country for deploying AI capabilities for national security for defense? Yeah, so when it comes to American civil liberties, there's a very robust debate historically, especially after 9-11 and Foreign Intelligence Surveillance Act, the National Security Act, 1947, all these acts where the government has tried to balance civil liberties and the good news has been like a robust debate on it.

And maybe those laws and rules haven't been updated yet. Maybe they should be. But if we don't trust that process and we're like, well, the laws are behind the tech. So I'm going to make a decision that impacts 3 million people in the department and then 350 million people in the country. You don't get to do that if you believe in the system. If you don't believe in the system as important, perfect as it is, then what do you believe in? Right? Then you're not, you're taken upon yourself to

kind of be God. And that's not, that's not something that I want, even though I'm a small government, free market person, you still have to have, the government has to have a monopoly on violence to protect its country. America is just as an idea, but it's also a nation. And it can only protect the people if we have the best tools and use them lawfully. And Congress is responsible for dictating that law. And we're responsible for writing regulations on that law. And, you know, we've got 40-page internal directives that have been there for years about autonomous weaponry. And we're looking at Ukraine and Russia and seeing what's happening there. And then we're looking at the Chinese stealing American models, taking the guardrails off, and potentially using those against us.

am I going to have my arm tied behind my back against the same model that has been stolen by the adversary? You get into Orwellian situations where it's hard to have anyone make sense of it and sort of be on the other side of that question. Yeah, absolutely. I think the point about our adversaries is well made. They are not having this debate. They are full steam ahead. How do we – was there anything more you want to share about where we are today in the AI – Where are we today in the AI debate discussion?

And like, is this put to bed or are we, is there a long road ahead? I think there's still work to do. And it's interesting, you have four companies that you could call frontier companies. And then you have this sort of major league baseball roster of researchers, like a thousand who they're trading amongst themselves, who, if you ask the leaders of these companies, are so incredibly valuable that it's a very strange dynamic for companies, a thousand sort of researchers who everyone feels are vital.

And how do we work with that dynamic where we have enough companies engaged with us? I'm never single-threaded again, which is a terrible gift that the last administration handed us. So we have multiple avenues who are interested in national security, who are patriotic, very much like the 2018 Maven where Google didn't want to bid on the contract. Now Google is a great partner and so I'd hope

some of the newer companies would learn from what Google did did, which is they didn't want to serve the government because they had employee mob issues but now they're one of the government's best partners. We see the Google Project Maven moment in 2018 in many ways as a galvanizing moment

to actually wake up a lot of founders and builders and say, hey, wait a minute, I actually do want the government to have the best technology and best capabilities. I do want to build. I mean, I would say if you fast forward, that gave rise to the American dynamism movement. We provided a landing zone for a lot of those founders that said, wait, I don't agree with the protesters at Google. I want first responders and, you know, the men and women who defend our country

to have the best technologies and capabilities. And whenever someone asked me these questions, I always remind them, look, like, Project Maven was used to make sure that we didn't leave any. behind in Afghanistan. Like that's literally the technology that was used in this public. And like, do you not, did you want to leave someone behind? Like, what's the alternative? And so I think that this will also be a galvanizing moment. I certainly hope it will be for a whole other set of founders that are excited to support their country and the people that serve the government. Let's talk about the Department of War structurally inside,

things that need to change, you know, in order to make sure that you have the best capabilities, best technology, and you get them out in the field, you know, we saw on the news, I won't ask you about it directly because you probably can't even answer. But, you know, we saw photos on the news of what looks like new technology, new capabilities. But what is changing culturally and structurally inside the Department of War? You know, obviously it starts with bringing in people that understand technology like you. But what is changing to actually enable the Department of War to modernize? Well, Secretary Higsas says we're on an unstoppable battle against the bureaucracy.

That's not the people. It's the bureaucracy that's built up over decades that prevents new companies with new technologies from getting their concept or their product deployed in the department. So what I'm trying to do with the various tools I have is trying to create normal contracting processes, normal requirements for example, we used to do things like, here's the thousand requirements in our RFP, a vendor would sort of fill it out and say yes, yes, yes, s, s, even if it was physically impossible, the physics didn't work. Right.

Then we put them on a cost plus contract, and they're like, oh, that didn't work out. And change order, change order at least for another three years of development, another couple billion dollars. So we're trying to move it to simple requirements. I need a missile that goes this far in this environment with this payload, etc. You industry come in me with your ideas and how to do it. And then you're going to five different approaches. Yeah.

Yeah. And we'll buy it and firm fixed price. And if you make a better margin because you're able to economize, this is sort of the Elon model, why he was so successful at SpaceX. Well, then everyone wins, right? And the venture community sort of is very comfortable with that model, right? You bet on winners. Some are not everyone's going to win.

But cost plus, endless development cycles doesn't work. And so we need faster development cycles, risk sharing with industry, clear demand signals, simple ways to do business. And those are the kind of bureaucracies that like every day I'm moving the debris out to try to make it happen. That's good. Like regulatory Moses. That's good.

So that to me is one of the most important things. it's changed and it actually gets the founders fired up to build. What is it that the companies can do better to work with the government, to work with you? You know, you've done this Arsenal Freedom tour. You've visited a bunch of startups. But what is it that you need now that the government is saying we're ready to do business, we're going to make it easier to do business. We want the best capabilities. What do you need from the companies and the startups? So we need companies to, this is, This is just the one thing the primes have an advantage on is not the inventiveness. It's the production and manufacturing ability to scale what they do with what they've initially built.

And I think the opportunity for startups is to take the great technologies and concepts, build up their muscle on how do you build a factory and produce these things that scale and do the quality test? and all that thing, all that kind of stuff. And so you do have to borrow from the old world to do that effectively, so you're not just reinventing the whole thing. And I think that's the next stage. And you're going to see companies start to cross that chasm here in the next one to two years where they've demonstrated or they've foot faulted, but then they've hired the right people

and they figured it out. And then that hopefully encourages a lot of other companies like, this can be done. And then venture capital dollars will continue to coalesce around. this marketplace, which I desperately want. So I think that's the thing to consider. And there are, the other thing, if I'm on your side, if I'm on the new entrance side, you could always find someone in the department,

three million people who say, man, this is great. I love it. I love you, man. I love your product. But money talks. So are they buying it? Are they testing it? Are they doing those?

Are they actually pushing you through the process? because that's the best signal of you, are you actually successful? Because the culture there, again, is to never say no. And I've tried to move it to faster yeses, faster no. So that if you're a startup, and I felt this pain as a startup guy, when you don't really know if you're getting a deal,

I'd rather be told no so that I can either move on to the next partner, readjust my product strategy, do something. And that's what we're trying to move the culture to. And what I tell the young founders out there. Awesome. Well, we're very appreciative of those efforts, and it's good for the taxpayer. Last question is we wrap up here. You know, you have a good life.

You're very successful. This is not a 9 to 5 job, I assume. You have a family. It takes you away from your family. What would you say to people that are in the startup and tech side of the world who are thinking about spending time working with the government, supporting the government, and look at you as an inspiration in somebody whose shoes they would like to follow and steps they would like to follow?

Well, I think, you know, everyone can, most patriots figure out a way to give back and there's lots of ways to give back. This was my way, which is, you know, I'm an immigrant. My first language was Arabic. I moved to this country. I get to Silicon Valley. I've seen some of the,

and been able to participate and build some of those interesting companies around. And it was the right time my life to say, like, okay, well, this is the way I'm going to serve my country. I want to have my kids see it so that they can see that. this you know our system doesn't come for free people they have to have builders people who care people who sacrifice to do it um not to mention all the board fighters who sign up for the jobs that say they sign up this is the way i could sign up and and you know put making sure that they feel that for me that's incredibly motivating and ever hope everyone else who's in the star of world finds their

own moment to do that if not in business but after in the right point your career and all that because We need more people to do that because we used to have government service as an honored profession, the Manhattan Project, the best scientists. Now, industries has a lot of opportunities to really exercise your brain and you're, you know, in those ways. But we can't forget that we need patriots to kind of come do these kinds of things every now and again. Well, thank you. We are so appreciative of the work you're doing. And thank you for being with us today.

We know you are extraordinarily busy. Thank you. Thanks for listening to this episode of the A16D podcast. If you like this episode, be sure to like. comment, subscribe, leave us a rating or review, and share it with your friends and family. For more episodes, go to YouTube, Apple Podcasts, and Spotify. Follow us on X, A16Z, and subscribe to our substack at A16Z.com. Thanks again for listening, and I'll see you in the next episode.

As a reminder, the content here is for informational purposes only. Should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see A16Z.com forward slash disclosures.