Loading summary
A
Ike Harris, naval officer turned Hill staffer who was most recently the tech lead for the RS on the China Committee and is now the executive director of the newly christened Frontier Security Institute. Join China Talk or like sort of war talk, I guess, because we got Tony Stark co hosting. We're going to talk gain AI Overwatch, the Hill's role in China technology policy, as well as Tease. Why the Penta Tease's new organization, which is focusing on a lot of cool national security AI evals deployment type stuff which is under appreciated in the discourse but is really necessary, I think, to actually get the most use out of this technology. Ike, welcome to chinatalk.
B
Thank you very much. Good to be here. Thanks for having me.
A
All right, well, let's start with Congress's role. Congress. Sorry. Well, let's start with this Congress's attempt to really direct export controls in a way that, I don't know, at least since the Cold War they haven't tried to put the squeeze on the executive branch. What are the Overwatch and Gain AI acts? How did they develop and why are they interesting?
C
I'm thinking about, I have a lot of friends in the tech industry in grad school, PhDs who are like, you know, maybe it is time seeing all of this that I get into tech policy. Maybe I go work on the Hill. I'm a big proponent of anyone at least doing a year on the Hill. How do you become a psm? How did you fall into your role? How do you kind of recommend tech people who want to take that as their next step, go after it?
B
Congress needs people that are both, you know, obviously intelligent but hungry because it is a very, very high paced. You know, you're always underprepared, you're always, you know, lacking for attention. There's a ton of voices that are trying to get that attention. Not all of them have, you know, they're, they're. Well, obviously they're all parochial, right? Some, some are parochial in that they are interested in a company or an industry. Some are parochial and that they are interested in national security. And then when you're dealing with the, the government directly, know there are interagency fights. So you know, you're never really, there's never really a clear picture and it's always a sort of a do, do the best you can with as little as you have. So you know, tech industry experience I think is a valuable, valuable asset. But really, you know, going into the executive branch or the, or Congress, I think is a useful endeavor for anybody that wants to be engaged in policy because you really get to see how the policy is made to understand it as a psm.
A
What different resources did you wish you had? Were there other sort of structural dynamics? Is it all just like, oh, we're polarized so nothing can be done? I don't know what's the optimistic vision for a more effective Congress broadly going forward? And then maybe we can talk about the China specific stuff.
B
Laws are the product of compromise, right? No law, that's very few maybe, and I would probably almost say no law is ever written and implemented the way that the PSM or the author kind of drafted it. Once these things become public, they go through multiple committees, they go through different houses, they get industry lobby, they get the government feedback from the executive branch of which multiple executive branch agencies may have different opinions of one piece of legislation. And then all that kind of goes into a thing that may or may not get passed. And when it does get passed, it's getting horse traded on the floor about specific things that one member wants to get on of another thing and get that onto this whatever. So, you know, these things are the product of a very messy system and they've seldom come out like this is the perfect bill, you know, it's just not a thing. So if in that, in that ecosystem, I think the companies are very reticent to just say like, yeah, we want this to happen, it's a good idea, go get after it. Because they know that the, the sort of devil's in the details and those details are very, you know, can be very hard to predict how they will play out. So I think that that's a bit of a barrier because as much as I would have liked to have just a, you know, hey, we're, this is, you know, this bill, you know, whatever it is, is in your long term strategic interest. That may mean you have to take a bit of a haircut now to, you know, to your revenue. But we think this will be better for you long term. They are often skeptical of the long term. What does any government official say that's going to happen after they leave? They don't know. We're all one election away from the entire thing changing and going different directions. So I think that as much as I would like that to have changed, I think that's the nature of our system.
A
Reflections on your experience in the Navy and what it did and didn't prepare you for spending time more on the, the policy side of things over the past few years.
B
Yeah, sure. Nobody's ever asked me that. That's A good question, Jordan. Thank you. I think the, you know, getting, having yourself grounded, particularly in your, your formative years and, you know, this, you know, the tip of the spear, so to speak, on, you know, people that are volunteering to serve their country for, you know, a job that doesn't pay particularly well, that puts you away from your family for large chunks of time, very unpredictable in conditions that are not ideal. Let's just say you get a respect for the people that choose to do that, that those of us that are making policy in Washington should really keep that in mind. And I think oftentimes the budget battles of the Hill and even in the Pentagon and sort of these, the different policy fights and all the different DC stuff can oftentimes in the national security aspects lose the connection to the guy, the girl that's, you know, in an aircraft, on a ship, you know, on the ground somewhere, somewhere that, you know, they didn't choose to go. They volunteered to sign up and then after that everything else was taken from them in terms of their freedom of action. And they're just trying to do the best they can to, you know, to defend the nation. So I think having that for me, that very, you know, visceral connection to that, that experience. I did all of my shore duty in the Pentagon, but all I, you know, I kept going back to sea. So I, you know, kind of had both, both views of this DC apparatus that is just, we're the largest bureaucracy mankind has ever built. But going to the very edge of that system, that's DCs at the top, the Pentagon, and then going back out on a cycle was good to keep perspective. And I think I've tried to maintain that even now in my last year in Congress as we're making, writing legislation on, you know, chips on, you know, whatever it was, you know, we, we passed, we got biosegure through last year. We got the COINS act, the outbound investment regulations. So, you know, those things are, are national security is done for a reason and that reason often is to protect the guy who's on the front line. And, and having that direct connection I think is extremely important and I, I am very thankful that I had that experience.
A
Tony, you want to, I'm curious for your take on the veteran to policy wonk pipeline.
C
Yeah, so it's, I, I think it's interesting because I've really seen a couple types. There's folks who want to solve a very particular niche that they saw while they were in service, but then they kind of get stuck in the field, and they're like, I don't know how to reinvent myself. And then they follow something that they don't really fully understand, right? There are the people who are like, this system needs fixing. I'm going to go fix the system, or there's a particular mission, right? So like, I got into China policy and I was like, that's. It's not one niche thing. It's all of the things underneath it that contribute to the success of that. And then there's the third one who are just really politicians and they don't really have a cause to go after, but they know they're there and they rely on that veteran image, various degrees of success for all things. Something that I saw repeatedly, and I'm curious on your perspective on this, is that there were definitely a lot of veterans who came onto the Hill, more officers than enlisted. But I would say both that thought that their veteran knowledge was the sum of all knowledge that they like, that was. That was their contribution. They did not need to learn all of this new stuff because everything that they had learned had been the sum of what needed to be learned. I'm curious how much you saw that.
B
Yeah, I think one of, you know, I've had other people, you know, junior officers as a commander, you know, how do I do what you did? How do I get to these things? And there were no. There was no, like, plan for my career. And I was lucky enough that in the service warfare community, my naval community, that, you know, most of your performance and your upward mobility is judged at your time at sea. And you have a lot more flexibility on your short duty than in some other communities like aviation, where you have to go to the training command or whatever. So I was able to do a lot of different stuff. I was in OPNAV and Pentagon and the Navy staff for a few years. I went to the Hill as a Congressional fellow. I was in the Joint Staff. I went to osd. So I think kind of to your point, I had a lot of different experiences. And the two things that I took away from that was one, like, no matter how important you think you are or how, you know, integral to the system, you will leave and you will be replaced and nobody will miss you, right? Or maybe, maybe for a couple of days, like, ah, it would be nice if Ike was here. But like, that ends extraordinarily quickly. The machine just turns over. It's designed to do that. You're not that special. So that, that is, that is good grounding yourself. And I had that Experience very much in my first short reading, the Pentagon. The second thing is that that turnover gives you some distance, some perspective that you're going to do this for some period of time and you're going to do your best and you may fix it, you may fix some of it, you may make no progress whatsoever, and then you're going to leave and it's going to be somebody else's job. So there's a lot of benefit to having multiple experiences, particularly in the government, when you're trying to do policy, to understand when you're in the Navy and you've never seen anything else but the Navy. Like the Navy staff is, you know, is the pinnacle of human existence. And then you go to the Joint Staff and the OSD and you're like, oh, the Navy staff is just one of multiple staffs. And then you go to Congress, you're like, you know, Pentagon is just one of multiple branches of the executive branch. So, like, having this understanding of where things sit in the bigger picture and DC is very much the big picture, and then how those things all work together is extraordinarily useful because it gives you an understanding of how they do or do not get things done, or what are the sort of cultural aspects of the integration of that of Congress and executive, of the individual interagency perspectives. And then even within an agency like the Pentagon, how the services and OSD and individual secretaries or Assistant secretaries work together and all that friction is how we make policy. So I 100% agree with you that having guys that served on the front line, you know, that did an operational tour, and then they go to the Hill and they get out and that's it. Those guys bring a fantastic perspective, but that's often a very limited perspective on their specific experience. And then applying that is like, oh, this is the way it works is I was on a, you know, a Marine platoon for two and a half years in Iraq, or I was on a ship and in the Pacific, or I flew an aircraft. Like, you do that for four years and you get that perspective. And that's one thing, but that's only one thing of this huge machine. You got it.
C
Yeah. And so kind of to that note, you mentioned, like, osd, and obviously OSD and Congress traditionally have a lot of communications between them, some hostile, some less than hostile. Right. But having seen both sides of it, like, I was always, it was always kind of funny going from Congress to OSD of, like, people, like, the common thing that you do on the Hill, hey, write a letter just asking for X, Y and Z or so. And so staffer just wants a briefing on this, you know, and the panic that it would set off for the most mundane programs in osd. Like what, what on both sides? What do you wish staffers knew better about OSD and that OSD personnel understood more about Congress?
B
Yeah, I think this is, you know, I think from the staffer perspective. Yeah, you have no idea what kind of dread you inspire when some challenging question that there really isn't a good answer for, you know, that somebody's got a, you know, it's a Navy question, but got to make sure they don't get in front of OSD or in front of the secretary or whatever and, and how much they have to make sure everybody gets a chop on that and what is it implicating potentially so that the. Yeah, I think there is not a good appreciation for oh, I know this guy at, you know, at Ledger fairs and Navy, he's a good guy. He'll just answer this letter. It's like, oh my God, that's like 45 people spending like a month of full time work to like get you a very simple answer because maybe a bunch of people know part of that answer but nobody knows the whole thing. And making sure when it's put together it doesn't say something that is against or the sort of is ground truth, I should say. And then the other side, for the most part, I think the Hill, particularly in the armed services and trying to select the hpsi, the other sort of national security facing committees. It is largely a group of people that are trying to, to help the executive branch do a better job at their, at their job and to give them the resources they need and you know, you know, get rid of some, maybe some bureaucratic hurdles that exist. So there, there is, you know, and this is the best of times when you have party one party control of both Congress and, and the exec, you know, I, I spent some time, I did my, my first Hill time was in 2014 working in Joe Courtney's office as a, as a Navy fellow when it was a Democratic president. But the Republicans had both houses, I believe. And everything just gets, you know, tainted with the politics of the parties and you know, the staffers have the ability to have a good conversation, I think come to some, some good outcomes. But you know, at the member level and at this, you know, sort of political level and then agencies, it becomes, it becomes a political fight because it is so it's tough.
C
It's funny because the other half of it is that there's definitely a lot of, you know, we get to the personal office. PSMs are their own beast. Right. The personal office is the new folks who work as MLAs or whatever if they've not worked military before. And I think I'd probably get guesstimate that like 80% of MLAs are probably not former military or at least not former Pentagon. They see the DOD as like this mystical black box of like, surely they have the answers. And so if there's anything staffers can understand, it's that, like, hey, like, if you're not trying to play gotcha with a particular office, like, understand that there is not this, this world of perfect information that comes from the DOD.
B
I 100% agree. Yeah. And they're, you know, my experience outside of the sort of ledge crowd inside the building, inside the Pentagon, you know, the average military officer doesn't have a good appreciation for the difference between a PSM and a personal safer in a committee, particularly if it's a, if it's a personal office staffer coming from a member of their committee. They probably would not pick up on the nuance of that in a way that, you know, would provide context to that.
A
Let's do the lesson right here. Why don't you do our little.
B
Sure. So I would say generally, personal office, and this is a very general statement, and I could pick out 100 examples myself where it's not true. But generally speaking, stuff coming from personal offices is largely about constituents. You know, the member is there because he is representing his district and the people that elected him. You know, a lot of the Armed Services committees, those guys come from districts that have heavy either military presence or manufacturing capability for the military. So they're largely talking about stuff that matters to the people that voted for them. On the PSM side, it's much more issue specific. So, you know, if it's, if it's the HASC staffer, they're talking about something that has to do with probably the ndaa, Right. Because that's the main bill that HASC staffers work on. So, you know, they're in some subcommittee and they're working on some piece for the ndaa. And the member, the same member that asked for that personal, you know, thing about, you know, plane construction or some thing that goes into, you know, produces some jobs in his district that goes into the military somehow may ask a completely different question about some very substantive, substantive matters have to do with the NDAA through his PSM that his staffer would Ask something much more district based. So, but, and knowing, you know, knowing who they're getting it from is, is part of that, is part of the understanding. And, and you know, Ola, I think the Olas, you know, we'll give credit to these guys, you know, if, if they're either, you know, on the Hill as the liaisons or back in the Pentagon is the Legend affairs folks. Those guys do a pretty good job of keeping up to date with this stuff and understanding the context. Most, a lot of them, maybe not most, but a lot of them have been congressional fellows. They serve some time on the Hill in an office, so they are fairly good at this. But a lot of times these questions, they're not answered by Legend Affairs. They're sort of handled by legislatures. And that's where some of that lost in translation can come through, I think.
C
What was your favorite, like parochial fight that you had to suffer through? Because there's two off top of my head for my time, which was like the lumber versus concrete lobbies for who gets to build on DoD facilities. Because those guys were like, we normally never talk to you. We would prefer we didn't. And then the NDA debates over like Made in America anchor chains where it would just stop all other progress on other things.
B
Yeah, I once made a LinkedIn post about the Jones act and that was a mistake. If you're familiar with the Jones act, basically it says all American. I'll probably butcher this, but it says basically any American good. Going from American port to an American port has to be an American ship that's crewed by Americans. Maybe news to some people, but we don't crew most of our ships. Most of the stuff that comes in the United States is crewed by foreign ships that are foreigners, Queen them. But us, to us, it's a big deal. So anyway, it's got a lot of different interests baked in that. So that was always kind of an interesting third rail. This is, I think, long enough that hopefully nobody would get upset about it. But back in, when I was on the hill in 14, we were fighting over the cruiser decommission while. And the administration, the Navy had kind of bought off on it generally was going to retire the cruisers or keep the cruisers around. I think the Navy wanted to get rid of them to get to shed the manning. And Congress was like, yeah, but you got to tell us what the new plan is. We really have a good plan. And like, yeah, we're going to build more destroyers, but we'd kind of settled on, you know, dealing with whatever Congress wanted at that point, the Navy had and Admin wanted us to go up the Hill and the Air Force was like getting rid of the Kiawah or maybe the army was getting rid of the Kiawah and there's all these like. It was just purely a political move of, you know, the services had to go up with a hill to die on so that, you know, the Democratic services had to go up with the Hill to die on so that the, so that they could die on the Hill. For the Republican controlled Hill at the time, it was, you know, purely politics and the service, which is, services were just kind of, you know, stuck in the middle, I think. And it is often a position they get stuck in and I. They do not like it.
A
How does all of this add up to the Frontier Security Institute, Ike?
B
I started working some AI policy around export controls, particularly around the chips in 16, and then again a little bit later when I was an OSD as well as reviewing CFIUS cases. And as we started to craft the outbound regulations, we started to look at how we could do that for AI. And then with the Hill and I thought I was going to have a very different year up there. But the chip debate kind of reared again and I think again part of that was the deep SEQ moment, if you will, that AI sort of became a lot closer than most of the experts thought it was. So where I had a lot of executive branch experience kind of doing this stuff, I was now in a position where having to make the, the case of why these export controls mattered, why, you know, giving chips to China was a potentially a problem, at least a balance of whether or not you want to continue the revenue or, and then maybe potentially lose some of the capability to China or at least advance their capability. So all of that was, you know, 2025 was a very big year for chips, I should say. Chip policy and AI so spent a lot more time on it than I had anticipated. And you know, as we were looking at how this, how AI sort of integrated into the national security establishment generally and the Pentagon and how we were implementing it and all these other things, how it's being used, we got, we sort of realized that the, there is a gap in the understanding from between the government, particularly the national security side and the labs that they just don't speak the same language. Right. The national security side speaks about operational objectives. We want this thing to do something. AI is much more engineering focused. It has a capability of some technical parameter and it can do many things. So I Saw very specifically that there was this missing translation layer of basically somebody that could speak a little bit of AI, which I was able to do poorly and then have the perspective of the operator along with, you know, some experience in the building and understanding how, you know, the regs were made and how the laws were made and all this and that, that sort of those things combined were I think were, I was a little bit of a unique position of having all of that experience at the same time and it being, you know, relevant. So that's where FSI was born of and helping, you know, as from a non profit side, helping the, the companies, the labs and all the surrounding commercial ecosystem develop the technology in a way the operators will be able to use and understanding the risks of that, but also helping the operators and the national security side understand, hey, what is possible, how does the technology work? How do you have to change the way you do testing, evaluation, acquisition, reliability. All these things that you know, we've done for a long time but never done done it on a technology that actually makes decisions or could make decisions or helps human make decisions with, with novel information. So we saw a need for that regularly and I thought so I was leaving the Hill. I, you know, I love, I love, you know, I'm a glutton for punishment and I love being in the middle of these, these kind of tough problems. So I figured I would do my best to, to move the ball forward on helping it get better.
A
So one of the things I was really interested to see on the initial site was this idea of tailored AI evaluations for models. Obviously there's been an enormous amount of money spent trying to make reinforcement learning environments and have evals for things which are like, very clearly turn into revenue like software engineering. But what's the gap in the national security space that you're thinking about trying to fill?
B
Like most of the evals are largely based around some sort of like specific scenarios that you can define and say, hey, you know, do you, can it do this if it's you know, bioweapons, uplift, if it's, you know, cyber capability, whatever. And the problem with, you know, war and even, not even, you know, conflict, but just military operations is they're incredibly dynamic and there's no predicting what you will do when you, you know, leave your, your garrison or your, you know, your training environment or you depart from home port or whatever. So you know, having the ability to, to you know, sort of test and evaluate these things to get to some level of like it's deployable but then understanding that I have no idea what's going to happen when deployed, so how do I ensure that the system is operating in a way that is at least acceptable to my use case? And that use case is going to be extraordinarily specific to what I'm using it for, obviously. But a model that's helping you detect and classify electronic magnetic radiation from a seeker or from a radar is going to be different from something that's helping an operator discern hostile from non hostile air contacts on a radar picture inside of a operations center. So the way I kind of think about it is the military buys a lot of electricity, it buys electricity in a lot of different forms. Electricity is not something it buys. It buys generators, it buys batteries, it buys power off the grid. If it's at a base, it's very specific to what the use case is. And it's got to make sure it's measured and it's, you know, given the same way. And there's reliability and you gotta have redundancy and all this other stuff. So you can't just buy AI, you have to buy something specific, understand how it's gonna work in that, in that circumstance, understand how it's not gonna work in that circumstance, and then make sure you have all the redundancy built in to say, you know, hey, the humans aren't gonna be able to just are gonna throw up their hands when all of a sudden they're thrown back into having to do it because the AI went down. So there's just a lot of factors around the military's and national security. I see use of AI that just aren't in the commercial world. And exploring that more, defining it more is going to be something that's going to be critical to deploying AI at scale.
A
I'm particularly kind of fascinated by this question of strategic level evals where it's less the kind of like you can kind of envision like a targeting evaluation, right, where it's like, okay, can we recognize the thing well enough, you know, out of this many times, you know, how would a human do relative to it? But once you start getting up to like the operational and strategic levels, like you can't run the Iran war experiment like 10 times to decide whether it was a good idea or not to like blow these things up before those things or, or even not from a sort of blowing things up perspective of like, okay, you know, what is the AI lift that I would want in doing this negotiation with the Iranians to try to understand, you know, what their red lines are and where we can push. And you know what, we should even be prioritizing when it comes to, I don't know, whatever long term objective you're optimizing for. Do you have any thoughts on the sorts of paces you would want to put AI models to, to through to help with these questions? Because my, my, like, like I am sure that there are just overworked, tired people in all of these jobs who are going to be using these models in some way, shape or form, whether or not they have a good sense of just to what extent you're going to want to sort of like plug them in or substitute your judgment for theirs.
B
Yeah, I think, I think that's, I think that's spot on. Right. I mean and this is where something is so simple as targeting. You know, the way I think about it, you know I had geeks, Mike, global command and control system, Maritime, you know, on my ship other than which filters are applied and this is just a, you know, basically a big two dimensional screen that shows all the units. That's basically it, friendly and hostile. That looked the same on my ship. It looked as it did in the tactical level or operational level at the carrier as it did at the strategic level at the CoCom. It was the same system, maybe different user defined interface or filters or whatever. But otherwise it was the same thing. Right, like targeting at the tactical level is looking for units.
C
Right.
B
I'm trying to find the tank, I'm trying to find the ship, whatever in a sea of an ocean, maybe a bunch of people. If it's looking for individual at the operational level you're sort of like what do I need to hit to achieve the objective that the strategic commander has given me? So how do they generate force? And then at the sort of strategic level what do I need to influence to achieve the war end objectives that the President's given me? And all of those things are incredibly different and taken to a, I mean just entirely different training set around how does one country generate military power versus what is its individual military power versus how it's sort of national leadership, applies control to its military. All these things are vastly different. So I think one of the ways that I could foresee this happening is looking at it differently and saying I'm trying to define at the strategic level what I'm trying to do at targeting, what I'm doing at the operational level, what I'm doing at the tactical level and then say if I can define it as I'm going to put pressure on a country or on a military or on a unit to develop or to, to achieve this end, be it, you know, eliminating unit, putting, you know, stopping the military producing more units or, you know, compelling a country. What are the things in that country specifically that I have to go after? And I think that would be, you know, it's, it's a, it's a multi, you know, multi variable, complex problem. I think those are some places where through both an understanding of just a vast amount of data on that country and then maybe some modeling done through AI, you could actually start to discern maybe some pressure points that are, they're just not apparent before having that level of aggregated data and analysis. And then how you would then go about using AI to then put pressure on those points in a way that was most effective. I think that's, you know, that's not removing the human from the loop or doing anything that would change the use of force, but it certainly may make your use of that force much more effective and much, you know, potentially limit conflicts because you can achieve it faster because you know what you're going after.
C
So I'm a bit curious, just building like, you know, the test cases and such like the US leads in AI, clearly, especially operational AI. Is anyone else like doing this or are there labs in the uk, are there labs on the continent that are, or even government organizations that are thinking about how to do this or are they just taking the lead from us?
B
It's tough to say. On the European side, I think there's been some great work by Jamestown and CSET on how the Chinese are doing it and actually pulling what they've been able to find from contract data and different stuff. To look at where the Chinese are buying AI or at least putting out proposal request proposals for AI capabilities that gives you some indication and it is vast. They're doing it for everything. So I think at this point we're still in the sort of experimentation phase. I can sprinkle AI on the things that I'm already doing and just make it better. Like, you know, you've heard now about this increase of, you know, from 100 dimpies a day to a thousand to 5,000 in just in the Iran operation. You know, that was just, you know, around the edges on the targeting cycle. Right. That's not a novel use of AI, that's just making what we already do better with AI. There's things we haven't even thought of yet that that will come from this capability. That is, it's going to be challenging because we're developing AI in a very different way than the Chinese are. We're looking at these big models and APIs almost exclusively through closed source of the advanced models. That's tough to do when you're on the edge. It's tough to do on a platform. Can't put a data center onto a B2, it's too big. But when you're using a deep seq or kimi or some of these other models that are near frontier capability, but much, much smaller, you have an ability to put them at the edge with the weights. They're open source. You can do some specific training on them in a much easier way than you can with the structures that our system has produced. And I think that we don't yet know what that will look like when the Chinese start to implement AI for, for the military. And I think we need to be very, we need to put some, some chips on. You know, just being imaginative on how they're going to do that so that we're not surprised when it happens in combat.
C
AI race aside, there's also the hardware production race. And so, you know, understanding that the, the PLA is not exactly a duplicate of us, and there are things that they value in terms of their mission set more than us. But, you know, the day that they start printing UAS for themselves, the way that they, you know, the, the way that the Ukrainians do or the way that they do for the Russians, I mean, that's, that is something that, you know, folks hear as much as they talk about. Well, you know, we built, you know, 300 switchblades or whatever. And you know, I mean, it's, you're gonna have to ramp it up even more to keep pace with that, or you do what the US Government's doing now, which is, well, maybe the solution is actually in defense. And you go after CUAs as the primary, or CUXs for all of them as the primary method to keep up.
B
Yeah, I had the, I was very fortunate to be able to go to Ukraine a couple times in, in 2024 and saw some of their frontline drone units. And it was a completely different model than what we would use. I mean, they have, they have battalions that are dedicated, just not, you know, a, you know, a, a private in a platoon that's like, he's the drone guy and he got, you know, five minutes of training before going out on his mission on how to throw the thing and then use the joystick. That's a little bit unfair, but not, not a whole lot where the Battalion size. I mean, these guys are. That's all they do is drone warfare, right? They supplement fire teams that are going out on missions. They have a lab that is near the front line that they go back. They can debrief after a mission and say, hey, this thing conked out within 20ft of the tank. I don't know what happened. And that guy can say, all right, well, this one had this waveform on it. So we know they're doing X, Y or Z. They can update that thing and they can turn around into a new system, literally freshly printed drone within a couple of days. I mean, that cycle time is dramatically different from the way our system works. And there's a lot of people like to talk about how we can. All the stuff we can learn from the Ukraine war and apply it to. To our systems. There is a ton of stuff. A lot of it isn't applicable. We're not in an existential war in our homeland that's, you know, you can drive home from the battlefield at night. We're gonna have to project power much farther and all the things that go with that. But there are some things around the tactics and the way we use this kind of new level of warfare that the Ukrainians are very, very good at, and the Chinese, frankly, are soaking up all of that stuff through the Russians. So they're implementing, I would imagine, faster than we are, just because their, their ecosystem is producing the drones. It's producing the, the, the data that those drones are seeing and, and it's cycling it right back into capabilities, going
C
back to the battlefield to that degree, though. I mean, the, the pla is very C2 focused, right? This emphasis on the commander being able to have control of everything in the battlefield. And there's always that debate in the US Military too, but we still believe in mission command. You know, is it possible for them to really kind of build out autonomy in a way that allows the junior operator to run on it, or are they just going to try to skip that phase to where you have this kind of universal controller, because they don't trust the individual to control a swarm.
B
It would not surprise me if Chairman Xi thought that he could use AI and basically eliminate the humans, that all they have to do is launch the drone from somewhere and that his direction will be given out via AI to the individual units, and he doesn't have to worry about, you know, disloyal PLA officers somewhere in the chain of command. Obviously, that's not our way of warfare, but I think there, that is something that they are trying to, to strive for. And I would, I would add that they are far more risk tolerant on those things going bad than we are. You can just go to Twitter X, excuse me, and search for Chinese rocket boosters dropping on villages. And it is a frequent occurrence that during their space launches a rocket booster happens to land in a village. That would happen once in our system and then there would be no more rocket boosters landing on villages because there probably wouldn't be a NASA. That's just a thing that happens in China. So their willingness to sort of suffer human casualties in pursuit of a party objective is just far higher than ours. And I am concerned how that could play out long term. If the PLA is able to essentially substitute large numbers of humans for robotic drones. Maybe the first wave of, you know, the Taiwan invasion is not, you know, hundreds of paratroopers, it's hundreds of robots, if not thousands, that continue to flow in. Right. And then, you know, if only, if only 50% of them work, but you have, you know, 150,000 or some, you know, ridiculous number that just, you know, walking around shooting anybody they see you don't care. If you're losing them at a high rate, then it does. You don't have to, you know, resupply them the way you would with a human military. So I think that would certainly be in the, in, you know, in this sort of objective category of the PLA is to say, like, how much of this stuff can I just replace the human with a robot and just allow it to get, you know, you know, blown up, but then replace it with five more because I don't really care and I have the capacity.
A
So on my quest to learn Ball, one of the books that's been assigned to me is the Masks of War, which is a 1989 Rand study that kind of like runs through the different branches and how they think about planning in the future. Look, if things are going to get weird enough where we're just having robot armies, you know, there's something to be said perhaps for like traditional quote unquote, like ways of war no longer being germane or just being sort of a blocker on the way to get you from today to the future. Any thoughts or reflections on bureaucratic stickiness and to what extent it's like helpful and harmful, particularly as we're potentially entering a period of rapid technological change.
B
Yeah, I think, you know, you can kind of go back to how both the United States and the Russians developed, you know, kind of long range strike in the Cold War, you know, our. My longest cruise missile, certainly for a surface target was the harpoon for the better part of my career and hell, better part of multiple careers before me because we had a, you know, our strike, long range strike platform was an aircraft. And we had an aircraft because you could see the target and you knew what it was and who you were shooting and you were fairly confident that you were not shooting somebody that shouldn't be shot. That's sort of like a fundamental premise of the use of military force in the United States. Like as much controversy over drones or some of the stuff, Iraq or Afghanistan. The guy on the ground was always trying to limit the loss of life. And war is hard and bad things happen, and it certainly does. But the core function is to limit the use of force to achieve the objective that somebody above you has determined to be a lawful military objective. The Russian cruise missiles and the Chinese that followed that, they were just fire and forget. I'm going to fire this thing 150 miles, 200 miles, whatever. I kind of know the Americans are over there. It's going to look for something big and metal when it gets there and whatever is there, it's going to go after, right? It didn't have a lot of like, you know, sensor to shooter communication to ensure that it was only shooting the, you know, their, the version of the bad guys. So if that's your sort of like paradigm that you're willing to just, you know, kill innocent people to achieve your military objectives, then you really have a, you know, a scale thing becomes a lot harder because you don't have to put super bespoke AI onto every military robot to make sure that that person that it is attacking at that moment is a legitimate combatant by the rules of war and by law of armed conflict. All this other stuff, you're just like, that guy is not one of our guys. He's not blue, therefore he's red, so you can kill him. That level of lack of concern for innocent casualties pervades the Russian Soviet doctrine and it pervades the Chinese doctrine. So I think we have a much more difficult challenge on our hands to then make these robots into something that can be useful in our version of the battlefield because they will have to be much smarter than what the other side's going to do. Not to mention the fact that they're able to produce a whole lot more than we are right now. But that's kind of always been our thing, right? Our thing is precision and just using the minimal amount of forces necessary Even when fighting, you know, more advanced adversaries. So we're pretty good at it once we get going. But it does take us a little while to get up to steam, and I think that's kind of where we're at right now.
C
Cool.
A
Tony, any final questions?
C
What are you reading right now?
B
Actually, Jordan, on your. The podcast you did with Jake Sullivan, you gave him a couple books at the end. And I am slowly making my way through, I think, To Rule the World, the. The Russian perspective on the Cold War. It's fascinating. So I'm kind of going back to this one too.
A
Social History of the Machine Gun. Social History of the Machine Gun.
D
Yeah.
B
You gave that one to him as well. That's right. Yeah.
A
That one's really fun and spicy. I have like a half written book review. I need to do it again or I need to finish it.
B
There's a great book called On Killing. It was written by a Marine colonel years ago. I always come back to about how much humans really don't like killing other humans. And it's really a sort of a. Unless you've read it or served in the military, it's a. It is a very, very interesting take on. On humanity in a good way. It turns out we don't like killing each other, which is a good thing.
A
Amazing. I think we'll end it on that. This definitely evolved into a war talk, but no problems there. That's right, Tony.
B
Export controls are effective.
A
Don't forget, guys.
B
All right.
A
This is great.
B
Thank you.
D
A Hill staffer one Tuesday with the markup nearly done, sat back at his computer and decided, just for fun, he'd fire a friendly query off to Bob at osd. A simple little question on a program, easy peasy, a paragraph and answer Bob by Monday, please, no more. He clicked the button send and went to lunch without a chore, quite unaware that on receipt the note had blanched the face of poor old Bob, who recognized with horror the embrace of 40. 45. 45 Pentagon people, 45 souls underneath the five sided steeple. The colonels, the councils, the Vice Chief's exec, the dep Sec, the interns, all hands on the deck. The chops, the equities, the action review. A month of long evenings for one staffer's adieu. 45 people, one innocent letter, and nobody, frankly, will know any better. It went to PA and to LA ASD for policy, to USD for R&E, who passed to OGC, who tasked it down to J3 OPS, who tasked it up to plans, who tasked it sideways to the Strat who washed his weary hands, who tasked it to the service staff in triplicate, of course, who chopped, assented, footnoted, and tasked back with great force to oust I oust ans the jcs the Air Force, Navy, Army, Space, and Marines for redress to DCMA to darpa to the comptroller's deputy to the desk officer for Taiwan, which was peculiarly irrelevant, but no one took the time to ask him why. To Public affairs, in case the pressure chanced to pro to the Inspector General, who was never anywhere, to anyone with email and a clearance and a chair, to anyone they'd ever met or hoped to meet or knew. The letter went to all of them and back and through and through 45 Pentagon dumb people, dumb people,
B
dumb people.
D
At last, in mid November, came the answer, fully chopped, 200 pages, classified, the rough edges all locked, delivered to the staffer's desk in Rayburn 22, where someone else now worked because you guessed was true, that ours had gone to K Street in the middle of July, where in our lobby's on the very program that's no lie. He skimmed the answer, briefly smiled and dropped it in the bin, then called his old pal Bob across the river. Bob, old friend, I have a tiny question. So when you draft a friendly note to cross the Tidal basin, recall the 45 who'll weep, whose souls you'll soon be erasing. Their palatons will gather dust, their spouses won't forgive em, but Bob will always answer you, that's just what we have given him,
B
Sam.
May 29, 2026 – Host: Jordan Schneider
Guests:
This episode explores the intersection of U.S. national security, AI technology, and congressional policymaking—zooming in on the unique dynamics between Capitol Hill, the Pentagon, and tech leaders. The discussion features deep-dives into how AI is being integrated in national security contexts, the nitty gritty of defense policy work on the Hill, and reflections on the new Frontier Security Institute’s mission to bridge government and tech.
[00:52–03:14]
[01:22–09:06; 05:05–09:06]
[12:20–18:28]
[18:28–20:46]
[20:46–24:24]
[24:24–31:38]
[31:38–36:32]
[34:41–39:22]
[39:22–43:16]
[43:19–44:25]
| Timestamp | Segment/Topic | |-------------|------------------------------------------------------------------| | 00:52 | Congressional export controls & tech policy | | 03:14 | Compromise and inefficiency in lawmaking | | 05:16 | Military perspective on policymaking | | 07:48 | The veteran-to-policy pipeline | | 12:20 | Hill–Pentagon communication quirks | | 16:27 | Distinction: Personal staff vs. PSM | | 18:28 | Weirdest parochial/lobbying fights | | 20:46 | The origin and purpose of FSI | | 24:24 | AI evaluations in natsec vs. commercial | | 27:14 | Strategic vs. targeting AI evals | | 31:38 | US vs. China, open vs. closed AI models | | 34:41 | Lessons from Ukraine, PLA autonomy, doctrine | | 39:22 | Bureaucratic stickiness amid transformation | | 43:19 | Book recommendations | | 44:44–48:05 | Satirical Pentagon letter saga (“The 45”), closing poem |
This episode is a must-listen for anyone interested in the realities of defense policy, AI’s real-world uses in national security, and how institutions adapt (or don’t) as great-power competition heats up in the 21st century.