Emergency Pod: Iran + Anthropic

emergency podcast, Iran, plus anthropic at some point, but we got to start with World

War III joining us today, Emmy Prabasco of C set Henry Farrell, Mike Horowitz as well

as Brian Clark. I think we got to start with Mike. I mean, this is our first like big

meaty American precise mass like campaign, right? I don't know whether I call it a precise

mass campaign. I think it's notable that this is the first the United States used a system

called the Lucas, which is America's first precise mass system, you know, less than $100,000

can go a couple thousand kilometers. Like you can definitely shoot it down, but you have

to try. And, you know, ironically, obviously reverse engineered from the from Iran, she

had 136. So like effectively using Iran's own technology against them. Now Iran itself

like copied some West German tech from the 80s to design the shade. So, you know, whatever,

I guess like what goes around comes around or like something like that. But the it's been

interesting to see in the like from just from a military technology perspective in the context

of the Iran operation, the mix of in some ways, what we would think about is American legacy

strike, you know, like Tomahawk missiles, like that kind of thing, along with some of these

early indications of emerging capabilities like the Lucas, as well as frankly, Claude

being in the mix. And like, who would have thought after the events of Friday that like

Claude would enter the chat so early? I think we should start at the strategic level.

Sorry. Yeah. We do that. You asked me about precise mass. And so I was talking with,

so there's this like, you know, paper bombing to win the whole story of the 20, a lot of the

story of the 20th century is like bombing people doesn't quite always get you what you want. But

the difference between bombing in 2026 and bombing in like 1943 or for most of the 20th century

for that matter is you couldn't just kill all the people who ran the country. So I was going back

to ask chat GPT like what their or Claude code actually, what the comparisons were of like killing

leaders, but not invading the country. And it was throwing me Jagurtha of new media, like the

Byzantines overthrowing like boyars and stuff. I mean, this is not a particular, especially when

you don't like have someone do the coup for you inside the country. And then like, is there to take

over it? I mean, this is a relatively rare thing in human history to like pull off an assassination

from, you know, a few hundred or a few thousand miles away. So I don't know, like, like, like,

like, where are we on air power now? I mean, we're four days into it, like, obviously TBD,

but curious for everyone's, everyone's takes on the theory of victory here.

I think you got to, like, you just mentioned, you got to have somebody to pick up the pieces

and run with them afterwards because if any good autocrats go to eliminate any potential

competition in the 21st century, it's not like the British, when we revolted and we had a bunch

of people that could take charge and they didn't bother to try to assassinate them in advance. I think

you're right now we've got this air power can be very effective at eliminating the leadership

potentially, but then you've got to have the civil society that can pick up the pieces or you've

got to be willing to put that in place with people on the ground. It just seems like that's still

going to be the missing element. Yeah, I mean, if you think back to like the OG paper or something,

argument was slightly more nuanced. I mean, it was that coercive bombing when you more

precisely hit targets can be, you can be capable of generating concessions from the target. The

issue is that punishment bombing, just like, like, let's go like hit like random stuff in the country,

like generally usually generates a rally around the flag effect and the targets, and that it was

hard to get really big concessions out of out of just bombing. What what seems to be different

today is the scale and velocity of what you can do and what you can do accurately. You know,

there are reports that Israel launched more than 500 attacks on the first on the first day. And

frankly, given the number of like weapons that the United States has, there's no way the US did

that many. But you know, this really illustrates the, this really illustrates sort of a changing

world. And but although if you wanted to make an argument that it's not changing, the Israelis

obviously yet again had exquisite intelligence, which they seem to have on every actor in the region,

not name Hamas. And like thus had the ability to know, all right, like all these guys are meeting

here, these guys are meeting here, giving them the ability to execute these these leadership to

competition strikes on the first day. The other thing is, you know, air power is can

do this kind of thing if you don't have any air defenses to go up against, right? You see in Ukraine,

there's the air power is not really cutting the mustard there. And the reason is because there's

air defenses that are preventing somebody from being able to operate unpeated in the air space. And

Iran's air defenses are largely neutralized. So you can fly around and launch GPUs at people,

which is a lot of what Israel is doing is using basically J-Dams on targets because they don't

have to operate from standoff range. You could do that in volume and really execute a pretty

effective coercive campaign and take out a lot of the infrastructure. That's very difficult

to take out with precision strike standoff weapons. I mean, you want to any strategic level stuff

before we go ahead? No, only partly because it's, I mean, this is all, you're a lot of questions.

I think I have more questions than I have answers. So I appreciate Mike and Brian jumping in

and attempting to answer. Well, you can throw it. I don't need to be the only one asking questions

also. You guys can shoot back and forth. I mean, it's unbelievable. We've never seen the United States

attempt a military operation of this scale with goals that were this in some ways in co-hate.

I thought I was going to say that word. In that, sometimes it gets regime change sometimes,

it's about eliminating the threat, whatever that means necessarily. In some ways,

what the Trump administration has done is break the, ironically, break the pottery barn rule

from the Colin Powell era. If you break it, you buy it. That is just not how the Trump

administration sees the world. And so they are willing to do things that no American administration

has ever been willing to do in the past because they don't feel responsibility for governance over

places that they bomb. So is it a good idea? No, I think that makes sense. But I guess the

question is, is that a good idea at the long run? I mean, Mike, you know better than I,

you're a real national security person. I am not. But there are umpteen different arguments,

articles, books, talking about how doing this kind of thing doesn't necessarily end up in a good

place, not over the long run. So do you think that this is something which is possibly going to

have benefits? Do you think that this is going to be a disaster along the lines, not of a rack?

But of something similar where we do see continuing, continuing problems for years,

perhaps for longer than that. What's the long-term strategy as opposed to the just go in there

and to reduce everything that's there per? I'm not sure the Trump administration has a

strategy beyond 2028. And so in some way, I mean, I mean, I'll say this Henry, it'd be a real bad

look for our political science business. If the Trump administration can like do Venezuela,

do this, and then Cuba is next. Cuba is obviously next. And there's no backlash. And they're like no

negative secondary effects. If you, I mean, just in that like everything we would expect theoretically

would suggest that, you know, instabilities likely to occur in these places, you have power

vacuums, you have, which increases your risk of terrorism, which increases, you know,

like risks of like, like militia like groups in these places, lashing out, and that like,

this would be very dangerous. The, it is possible, frankly, that we don't actually see that in the

short term. And we end up seeing that in the long run, but the, but the effects in some ways like

look disconnects and enough from the initial operation that we don't, that that maybe like the

Trump administration doesn't care as much or something. But like, yeah, like, look, you're,

you're 100% right. I just think they're not, that is, they are unconcerned with like instability

per se, you know, like increased risk of like a terrorist attack or something like that, the,

or at least at least based on my read. But what's the, I'm sorry, I mean,

no, I mean, should we also give air time to the like the argument for why now? Because I think that's

the thing that's pushing against, I, you know, concur with everything you're saying. There's,

we've sort of opened Pandora's box now. So what happens next? But like, I think it's worth talking

about why, why some, there's another hand here that is at least worth discussing. Yeah. I,

well, yeah, I mean, the, the argument would be Iran's on its back foot. This is an opportunity to

take it out as a military threat. If your is real, this is a terrific opportunity to eliminate

the potential for Iran being able to threaten you with missiles or even a nuclear program

down the road. And you may not care that much with the turns into a, a mess, you know, and I'm

maybe not a failed state, but not a well governed state. But if you're a cutter or your, the UAE,

you may not appreciate that because now you've got to deal with this neighbor across the way that's

got, you know, all kinds of instability and it's probably interfering with shipping. And if you're

cutter, you depend on LNG exports for a massive portion, proportion of your economy. You can't

have the straight-ahore moves closed off periodically like the Red Sea is or else you're going to

actually start losing, you know, some of your economic livelihood. So I think, you know,

for the Gulf states, this is not great. And I'm sort of surprised that they didn't push back more

on the effort to, to mount this operation. But they're going to be the ones that inherit it,

probably not as real. Yeah. And I, I mean, not to say that there aren't like, I don't really

understand where this goes either to Henry's question of like what, what's our goal here. But,

you know, then I don't think we want to gloss over that, you know, Iran didn't necessarily have

a great government to begin with. So that's right. Well, my question, if you, if you want to do

the regime change, don't you do this while the protesters are in the streets and like not when

you have the 30,000 of the most eager people dead? I mean, it's like a, it's time's a little tricky.

I mean, sure, we take time to like line up a military operation to like get all your assets in

place to be able to do, I mean, you know, like in, in any given, and I'm curious what

anything's about this, like in particular, Brian thinks this in particular, like at any given moment,

if you want to like launch a couple Tomahawks or maybe like maybe you'd send a special forces unit,

like sure, like you could do that. But if you want to sustain campaign, then you've got to array

the forces. I mean, this is also why like I'm not that worried about a US grounded invasion of

Iran, because like the forces just aren't there right now to do that. I mean, it would take months

to frankly align the like ground forces if you wanted to, if you wanted to invade. Sorry,

one other thing, there, there's an element of this. If if Iran goes down and then Cuba goes down

and you think back to the like end of the Cold War period and kind of the rogue states that we

used to that we used to talk about, then in some ways like the Trump administration kind of like

an end of Godfather one way is like going around trying to like knock off like eat like everybody

on the checklist for like a reset. And then I guess like what happens if we get back to North

Korea? Taking care of the family business. Well, what does it do to nuclear proliferation? So I

guess that everybody's gonna get nuclear weapons. Now are you kidding? Yeah. Like the majority of

the South Korean public already want to nuclear weapons. Why would they stop now? Again, you could

think all this stuff's a good idea like whatever. Like it just it like you can be totally fine with

it, but it's not I mean, yeah, of course. I mean, I mean, the old school way of doing this back in

the in the bad 80s or 70s would have been having the intelligence services start to establish a

center, you know, another power center that would be able to take over when this thing goes down.

You know, the CIA did this routinely in South America and Central America. It seems like in this

case, we use the intelligence services to find out where all the head guys are going to be and then

be able to take them out at once. But we didn't do anything to establish the alternative that was

going to rise up and take its place. So that seems like we've not really thought that through

because it doesn't seem like there's any discussion about who's going to be potentially taking over

that we would prefer to the current regime. Well, Trump had a line. He said like, oh yeah, I had

some some ideas of guys in mind and like, oops, we just killed them. So now we're on like

due number 50, which maybe we don't even have a case file on. I mean, when we're that down in

like the minor leagues here, the deputy minister of agriculture. Well, is there so? Okay, let's do a

little, or I want to come back to something that Mike said earlier about like the theory of like

if this works, if this ends up working out, like what did we not understand about the world?

I don't think it's that we didn't understand something about the world. I think this is all a

question of like how you process risk and make like assessments of risk. Like the argument

against going after Iran, for example, has always been that essentially Iran has chemical

weapons. They have lots of these long range drones. They have lots of different kinds of missiles.

They have like a bunch of USVs. They could they could shut down the state of Hormuz. And so

you have the risk of a big impact on like setting aside any like moral or ethical like should you

fight the people like whatever. The they have a like should you fight the government. The

you have like a parameter risk of escalation or that their air defenses function better. Like

or that they can unleash really substantial terrorist attacks in Europe, shut down the

strait for weeks like that kind of thing. And in some ways, it might suggest that those risks

essentially were like assessed at too high a probability. Or I mean, it could mean that those

risks were accurately assessed. And we just like got the like play. We got like the dice roll of

reality where like we didn't see those impacts. But it's also possible because like history is

long and like winds around that we end up seeing some of these impacts just not immediately.

Or maybe it'll happen tomorrow. I don't know. Yeah. Well, the. Yeah. There is that old Bismarck thing

about God loves fools. Drunks and the United States. I use that quote in my US foreign policy class

like last week. Amazing. The terrorism one, I think we should dwell on for a little bit. I mean,

we've had like multiple assassination attempts that have failed by the Iranians. But they were kind

of using the B team. And you know, you've had like or they were trying to like contract it out

and they ended up contracting now to some FBI agents who are now all busy like finding immigrants

and whatnot. I mean, that's a real risk. It's not super excited to see how that one plays out.

I had an operational question. Oh, let's come back to like was it about was it about how screwed

we are in the end of this effect? We'll get there. We'll get there. We'll get there. I promise.

Like the idea that you can kill off this many people and around still running around

firing off all these missiles and doing stuff. And this should this be surprising, impressive.

Like what is that? What does that tell us? They've been preparing for this kind of thing for

decades. They've got the infrastructure to support distributed missile launches. It's still

got a couple hundred missile ballistic missile launchers available and untold number of Shahed Shahed

drones that they can deploy. So they've been preparing for this and largely have distributed

their command of control, especially IRGC, which survives and was trained to operate in a more

distributed way where they don't really need contact with the headquarters to be able to execute.

I think that's the tricky thing, right? Is like you have the Iraqi army give up when the when

like when Trump like sends a text message to all of them saying surrender or be killed.

Like they're not literally all going to be killed, right? Which is different than like if the tanks

are rolling in from Kuwait. So it's just I don't know. I'm worried. You should be worried.

Yeah, it doesn't it doesn't seem to be an easy way for this to end cleanly. You know, it just

seems like it's inevitable that there'll be is going to it's going to be protracted. And the only

question is protracted in what way does it result in continued closure of the straight-of-form moves

and economic impacts and or does it result in continued ballistic missile attacks that

eventually start taking out things that we care about? Other operational stuff you want to go to

trying to Taiwan like? No, just like it I mean I thought it was I thought I was I thought I was

striking. I mean curious what others think that the it's like day one of the conflict and you are

already have articles showing up in like the journal and like the times in the post like the US

might run out of weapons soon. And the I mean you know like obviously the you know far be it for

me to not take this moment to describe again how bad it is when somebody fires a you know like

$50,000 shot at you and you fire a million dollar thing back to destroy it and how thus we should

be firing the $50,000 shots. But the the that that is not sustainable and Iran is you know if

you know like presuming Brian is correct in his assessment of Iran's ability to continue launching

or they could even like reconstitute some of that launch capacity over like a month long a

month long period then I mean you're going to be really drawing down stockpiles because the US

isn't just protecting the US is not just protecting the US Navy or US military bases the US is

also playing a role protecting all the Gulf countries. And you know like recall how like how upset

they got like they have been in the past when like they have faced like risk in this context.

And so the US like that's where there's I think a lot of pressure on the US and that

means that if this keeps going the US will have to pull in theory would need to pull some stockpiles

out of the Pacific and like send them over to the Middle East to be able to continue

intercepting Iran Iranian attacks at the rate that they're being intercepted. And I mean and like

that's a like I don't know that's like that's like kind of risky. Yeah it's that air defense

interceptor inventory that's the big problem is we're burning through those at a pretty high rate

because even if you're smart and you don't use them to shoot down the Shaheeds use your guns

or you use something else to take down the Shaheed you're still you're still using a lot of them

to take out ballistic missiles. And then the Shaheeds are used on all the soft targets that are

undefended because you can't protect everything at the same level. So these Gulf countries are now

having to come up with a way to defend against Shaheeds which they didn't have to before you

know they got to defend their shopping malls and airports against long range cheap drones.

Not to mention like all the naval assets that we shift over there that could have been doing

other things. Right. Right. Right. 100%. Great point. All right. Well Mikey gave us our transition

earlier. The story came out that clawed or sorry anything else on around. I feel like we only

get a 20 minutes. This is like. I mean we could like we could like tell what a wild it is

like forever but you you've also like collected a set of people who are like interested in AI stuff.

And so that's not that is true. It's so like it's like whatever I'll like I'll like sit here

and talk about like what it means for the defensive Guam or Kadina or something as long as you want

but like the is like look you you like what are we I mean I also I mean sorry one more thing

that I'll say that you could like user not use frankly is like the I do there's an element of

this in seeing American air defenses have to operate at scale against Iran where I wonder what

information we are now communicating over to China about how our air defenses operate and where

the like soft spots might be a conceptually that that could then you may be in for I mean we know

I mean look they the Chinese based super close attention to everything we do and like this will

be no exception and obviously the world has had a very close look at offensive US capabilities

you know throughout the war on terrorism period and so they're certainly well versed in those

which is one reason why they've been nervous about them because they're pretty we don't have

enough of them they're pretty good but like now they're getting a really good look at US air

defense having to operate at scale yeah I but that yeah there's a the flip side of that too is

US air defenses especially for the sea based stuff is worked I mean the ground based stuff is

work too so I think it's I mean it's a little bit so I don't know Emmys perspective on this was

it but it was surprising to me as a retired Navy guy that this stuff actually works when the time

comes to get you get shot at and you pull the trigger and it actually works that it defends like

you thought it was going to do our two million dollar interceptors should work against the shaheed

they worked against ballistic missiles too I mean well that's what's interesting is that you

know the stuff it works now it's it's it's expensive and it's overkill for a lot of these threats

and so what this has done is given these guys a lot of sets and reps to evaluate well what is the

right defensive system to use against the shaheed and they're not using SM2s against shaheeds

anymore they're using guns they're using other drones they're using jammers yeah so they're

using other stuff against the shaheeds now so they but that's so it so it sort of goes both ways

you've got both you've given a bunch of telemetry to China that they could employ in their own tactics

development but we also got a bunch of feedback that allows us to refine our approach that

otherwise we would have been doing this against China and probably not doing it nearly as well

so i think verans got a great point there i'd also add in the operational experience in the red

sea where we've learned a lot and and the number of times that i heard about what was happening on

the ships out there and thinking oh my god you know you don't do those maneuvers unless you think

you're about to die and and it it has taught folks a lot i think it's sort of the reps and sets where

the reps and sets have begun and and so we see them extending now but all of this is fantastic data

for China yeah but it also you know it's combat experience right so you got Chinese military

with you know now zero combat experience with the most recent purge going up against you know this

is more experienced for the us forces obviously not against a pure competitor but it's better than

nothing i was uh uh did know uh general can just saying joint a thousand times um but i mean it's

relevant right like this is we really got a lot of a lot of players in this at the moment

well this is job right he's the chairman of the joint chief he's got a it's branding right it's

because he's not actually in charge of anything so you're in charge of the staff you know but in

reality so if you're the chairman you got to highlight the joint nature of the operation so you can

call out the logistics so don't forget about those and then the family members i mean yeah like

this is good he's given everyone their kudos all right uh might give us the transition earlier

who do you think you think it was anthropic or the department of work who put the story out that

cloud was being used in this sorry cloud was being used in what in the around operation

operation and epic fury um i don't think it takes much to figure out that sent comm is using maven's

system yeah thank okay like they they they tell us every single time they can right

i would so i would tell you uh maven smart systems that all the command commands

clawed is integrated into maven smart system that is not to say that everything that maven does

is clawed there's lots of things that maven does that some of which has absolutely nothing to do with

AI and purely just moving data around um but sent com's probably the furthest ahead they're the

most experienced warfighting cokom they are the most experienced with maven smart system uh general

krilla you know he's sort the og here who was at the 18th airborne really did uh incredible work

there and goes to to sent com um and i think you know to the reps and sets conversation we had

earlier i think they've been working very hard to get us more as they possibly can and use it in

the most responsible way so i don't know what exactly they're doing i'm i'm very happy not to know

exactly what they're doing but um they are the most experienced cokom in this sort of uh the use

of mss and therefore presumably integration with clawed fair enough yeah yeah like i i think this is

i don't know who it doesn't necessarily need to be that somebody like leaked it on purpose this could

easily just be like somebody asked somebody at sent com and they like happened to say it like

frankly that that is often more likely from from my uh by viewpoint than a like deliberate

stratagem to you like get the information to get the information out there like it's just like

plus one on everything and he said it also um maven smart system i mean not to throw shade on maven

smart system god forbid but um it's a little clunky to use uh you know it's not you know it's it's

like your typical web thing where you've got a lot of menus you got to go to and you got to pull

multiple things down and create a workflow and so having some kind of AI tool to help you do that

is almost essential i think to be able to have it run at an operationally relevant tempo um so

i think that's uh there's an element of that too that these these planning tools because there's

just so many parts out there that you can pull together into a kill chain that if you want to

do that at any sort of scale and tempo you need something to help you do it i mean i would sorry

just to push it even further Brian like it wasn't exactly like our weapon systems on our ships

were a joy to work with um or manipulate or even understand um and so this is a step change in

proof place that's true yeah it's all yeah like what's the baseline like what do we what do we

what do we get it's the point so i think uh i think actually one of the things that's the super

interesting about MSS is that uh it's got lots and bells and whistles there's a there's so much

you can do it's just you know it's workflows off where with a bunch of data streams that you can

manipulate which is awesome and it's super intimidating and you can't just like sit down and

expect to be able to manipulate this thing like you're doing Gmail uh it just takes a lot of work

but so AI does help but in my mind the AI helps less with sort of user manipulation or in the sense

of like user interface then it does in and yeah and uh just there's so much data there's so much data

available it's extraordinarily difficult to make sense of it um i don't even know that you know

necessarily the clouds making sense of all the data all the time because you know the use case

there are questionable sometimes but it's really good at like writing that at you know the daily

report right and i'm like there's some really boring things that every day in a operation and AI

can be helpful in supporting people who are going to do that anyway and and then allowing you

just sort of helping the person not necessarily replacing the person so i'd like to do that workflow too

yeah yeah i'd like to jump in on that i mean because here's my sense of somebody who has

spent zero time in any sort of part of the armed forces whatsoever but my fundamental working

assumption with all of this is AI is fundamentally uh in its current form a bureaucratic technology

that is it allows bureaucracies and if there is one bureaucracy that has the biggest bureaucracy of

all it is the DOD it allows bureaucracies to do things which uh are more efficient than

traditional means of paper pushing uh summarizing information translating between different

amounts of languages different branches use all of these really mundane and but nonetheless

quite crucially important tasks and so i i wonder when we look at this big fight between the

traffic and the and and access how much of this is really missing the point that you know so i

think there are real issues here which is uh we do you know on the one hand you can see ways in

which these uh these technologies can be used to automate certain kinds of aspects of operations

which are highly problematic on the other hand you can also see ways in which they can be used

and this is clearly part of the story they can be used for domestic surveillance because if you have

a bunch of disparate data about individuals you know so they might have sort of have uh you know

they might have different uh social media services they might also have dating services you can pull

stuff together in ways that make sense but that this is not actually about whether or not we're

going to see terminator happening in five or ten years time this is about much more mundane much

more ordinary albeit uh crucial and sometimes pretty scary uses that the technology could be used

for so i would love to get you guys a sense on that because that's my sense on the side yeah

did we just lose Henry um okay um yeah i i i'm in violent agreement with your hand

yeah everything you said is it's just right there's so much of this that's mundane i'll give you

and one of my favorite examples of where i've seen an unclassified demonstration of something that

would could be used on the classified side which is um foreign disclosure so there's this super

boring task that has to happen where we have all this classified intelligence and you might want

to share it with the partner they may not have you may not want to give them the full story so you

might want to tell them hey we have aircraft in a particular area but we might not want to tell

them how many and so you can you know put together an LLM an agentic workflow workflow if you want

that takes the original intelligence then kind of runs it through all the different parameters

and the different guidance documents that these guys get and then come out with a sanitized version

of the intelligence super boring totally a bureaucratic task um and this isn't to say that necessarily

it always gets at 100% right and you should never look at it but like the task of doing it in the

first place can be so much more efficient and and um yeah if we could if we could really help people

understand that nobody really wants a terminator or the you know this warbot meme that's going

around on the internet is not I don't want a terminator none of us um I keep trying to tell people

you know military officers fundamentally like control uh oh you know just to to seed so much control

is um not really in their DNA or their training or their bureaucracy um but anyways I this is all

violent agreement with you and does that help you at all super helpful I would I would add uh I

grew there in the Emmy set I would I would I would add one distinction which is I would distinguish

between AI and LLMs in that the I and this is part of where I think the like challenge has been

frankly in the warbots conversation as much as as much as anything since the Pentagon is deployed

autonomous weapons systems for like 40 some odd years essentially um and like to and and and if you

use the Pentagon's definition of autonomous weapons systems that's true if you use phrases like

anthropics like whatever fully autonomous weapons like big people should nobody should ever use

that phrase um but whether like fully autonomous weapons systems or like whatever phrases like the

NGO community concerns and frankly they've probably been using autonomous weapons for many more

years because they're like wrapping in a lot more wrapping in a lot of like precision guided

weapons and sort of things like that and like this is created a lot of challenges in the conversation uh

I think because like anthropic is certainly correct that like the last thing you would do is take

take like clawed trained on the slap of the internet and like slap it in a weapon system and like

try like like like hope that it would like hit the correct target like anthropic is right that's

not ready for prime time but you could use use which is why you would use instead I think I've

super bespoke algorithm trained on a very like very bespoke data set that probably wasn't LLM

based and but would still be a autonomous weapon system or even like an AI driven weapon system but

the like the things you would worry about the risks some of the control issues that like Emmy is

that that Emmy uh smartly mentioned are very different in that context but like that nuance has

just gotten lost here I think I can keep going on this forever but um the other thing to mention

is that um does anyone know of a perfect weapon system I certainly do not

there are no perfect weapon systems they are all I mean it's a perfect weapon system

Brian no no I'm like um but um they're they're all flawed and uh and so we learn to operate with

flawed weapon systems uh and we learn when to deploy them and how to deploy them and so I

I don't think that while listen I really appreciate that we're having this conversation I'm glad

that people are interested in this topic I think it deserves a lot you know a deep thought um but

I don't think as we enter into the conversation people recognize how many fail safes the military

builds in to its process and how serious you know they're these are still human beings that go home

at night and want to sleep you know with a clear conscience as if they can um and so we put in you

know there's the technology can you get the technology right you know can we get it to the highest

level of reliability and uh precision that's one part of it then there's all the training on the

people so you don't get to operate these systems without going through a lot of training and then

having usually someone higher up the food chain say like yay barely you are allowed to push that

button or do that operation and then on top of it you've got as Henry pointed out this phenomenal

bureaucracy that we have really perfected over time that is building an all sorts of checkpoints

like you may turn it on only at this time you may point it in only this direction you may

you know the rules of engagement the battle app the battle orders that there's just all these

different bureaucratic things so we we put in lots and lots of process procedure to try and boil

down the risk of imperfect systems and that's not to say that we should just go running forth with

with any of these tools but just that we have to build up all of the fun doctorate and operations

training around it so the it might also be important to draw the distinction between the autonomous

weapons and the AI enabled command and control and planning function because I think

those end up being very different in terms of like what the capabilities are what kind of guardrails

you can put in and then to what degree we're willing to turn that over to the AI either LLM or

or an AI enabled system because I'll tell you in our work aiming we find that teams are willing

at some point to just press the I believe button that what the what the AI is telling them the

course of action should be is fine because the it just gets too complex you're doing your maven

smart system you know kill chain development and you're running out of time and you're like okay what

do you think I should do okay that's good we'll execute that kill chain so the autonomous weapon can

have all kinds of guardrails on it but if we've built a plan that is derived from some model and

we're sort of just expecting that it's going to work and it's not going to kill a bunch of innocent

civilians then we're we're not gonna you know that basically obviates any effort that we had to

try to make the autonomous weapon safer because we're our planning process is not safe okay can I

can I maybe push a a slightly different version of the Dario Amade story because okay here's my

sense of things and I I've got to say I do not buy into the Dario Amade vision of we are going

to be in a world where we have a nation of geniuses in a AI lab pouring out stuff five or 10

years from now but I think there's some real concerns that I think his ideas to some extent get

out and here my my my sense and my guess is Mike is the same I have complete face in a lot of the

military esols that the United States has created the same day that Hague says did this he also said

that he was going to take away the opportunity for military for military personnel to do

advanced degrees at a bunch of universities claiming that there was this incredible hostility that

professors had towards the military and my experience I guess Mike says the same is that if you want

to have really thoughtful interesting people in your classes by and large people I'm sort of

aware from the officer core haven't or you know they have a standard deviation more care more

principle more ideas about these things than most people do so they're awesome in the classroom

they're awesome in the classroom they're wonderful and I think that's a pretty universal

agreement that all of us have that said if we're in a military where Hague says is also saying that

you're some more or less we don't want to worry about stupid rules of engagement that that

makes me nervous and if we're in a world where as Amy says these technologies are fundamentally

imperfect there is a ton of slop associated with them you worry about the about the ways in which

differences at the top and how you do things might intersect with these systems in very

unfortunate ways and you especially worry about this when you look at the possibility for domestic

information and and the various you know a lot of this seems to be about access to domestic

information there's a lot that the U.S. military can do under the law answered to get around 12,

3, 3, 3 gathering I'm served together information from commercial databases and I will be frankly

nervous about some of the ways in which this might look two three five six years down the time

if this was not pushed back against I worry I mean I agree with you macro I think

I I've been very vocal that I thought that the Pentagon's been adopting AI too slowly for

a long time rather than rather than too quickly because I think the like risks have essentially always

been that the U.S. military would rest on its laurels and has been too slow about integrating

emerging capabilities and in in part it's because of like all of the policy and procedure and

process like most of which has nothing to do with AI at all and so I think the risk for the U.S.

is generally been going too slowly rather than too quickly and and frankly even in the Hegseth

era though I'm a little less comfortable making this argument at present that even in the Hegseth

era that they're all that like policy and procedure and whatever like still exists in ways that

make it fundamentally difficult for the reason that like Emmy suggested before which is like

incentives are actually aligned to have systems that work because systems that aren't reliable

like by definition don't work and and operators commanders and operators aren't going to want to

use them because they they want to use things that they trust like if they can't trust these systems

I mean which is why like the like both Emmy and I have done research on automation bias in one way or

another the this like phenomenon of over trusting over trusting AI and you know it's like Brian's

point about the like thing that you worry about like people just like hitting the I believe button

but like the if you if you trust algorithms more than you should given their accuracy like

you solve that with things like training and standard operating procedures and like it's

frankly good that in these war games that like people are like get confused and then press the

I believe button because then that tells you how to that like on the one hand like sure you'd rather

they not do that but on the other hand like here I'll tell you something to make you feel better even

the I've got a draft paper that I'm working on with a woman named Lauren Conn and a woman named

Laura Resnick a Samitan that compares a sample of West Point cadets to a sample of a US like the

US general public essentially like similar like age and education level where the West Point cadets

are substantially less susceptible to automation bias than the general public and we it I

think that the mechanism is essentially this the mechanism is essentially the training that the

military is giving people like not just in AI but in like we're fighting in decision making in

general which to like Brian and Emmys Point actually can make people more cautious at times.

So if I can I can't leave the automation bias point to to go I agree I actually did a

study with also Lauren Conn where we looked at you can compare the way the army uses the

Patriot battle Patriot missile battery to the way the Navy uses the Aegis weapon system to

generally these are very similar but what I found what you find interesting and this goes to your

point Henry that if if we're going to accept imperfect weapons which frankly we we have no other

choice then you need the bureaucracy to sort of address it so in terms of bureaucracy but the

Patriot the guidance that is first of all the who they staff a Patriot missile battery with is

a slightly more junior person they have slightly less training there and in the training documents

at least when you go back to the original ones that are unclassified it says basically like just turn

the system on and don't touch anything because this system is smarter than you and you can read that

in the guidance documents if you go to the Navy what it says is this is your responsibility and if

you screw this up it is entirely your fault I was trained under that yeah this is so true

very seriously now that said you know both sides have committed terrible mistakes

the the Vincennes incident and the Navy in instance they didn't trust the system the system was

actually correct and they said I don't trust the system and then they accidentally hit the wrong

thing and in the Patriot Fratricide you know they sort of stayed hands off because they said we don't

know what we're doing so there's no perfect answer here is sort of a sad story but but there are

bureaucratic choices that can be made and so if we are eroding the bureaucracy if we are eroding

test and evaluation or all the different things that have to come after you buy the weapon system

that's problematic and so I think and I frankly think that our operators don't love I mean to

Brian's point like if you drop MSS on an operator's desk they're going to be like okay like this

is complicated but if you give them a proper training and you get them certified they'll they'll

become more facile they'll be better at their job I don't want to miss Henry's intel point

which is not my area of expertise I've I've learned enough to be very humble about how

intelligence works I think but I do think there is a worthwhile conversation to be had in the

country about what what do we expect what is now our expectation in terms of the data that's

available and while we are concerned about how it might be used domestically and certainly I'm

I'm in that camp you know same data is being bought by China guys and it's not like it's not

available and so this is more than just a problem of how do we choose to govern the way we use

data by our government but how do we choose to allow our data to be to be shared and and how

honorable are we now in a way that we really weren't vulnerable before I mean the data was there

but you couldn't really use it until you had these new tools I'm sorry Brian yours is me

oh no I was good just gonna say you got companies like Vannevar labs and others who are

commercializing this you know harvesting of commercial data on our enemies and then giving it

to the U.S. government so the U.S. government has benefited from this availability of open source

intelligence and data and we've been using AI tools to harvest it so there's a legitimate question

to say how how much of that is going to be like U.S. information that's leaked over into somebody

else's network and now we're harvesting it for the purpose of our unilitary intelligence gathering

so similar to Pfizer right it's the same challenge with Pfizer that I'm going to

supine somebody else but if they're talking to somebody back in the U.S. I'm now essentially

spying on somebody in the U.S. but so it's something that we have to ponder and I think back to

Emmys point the only way to really heart you kind of keep it in check is to avoid giving so much

data to third parties that are going to be able to then buy it to somebody else can we come back

to the fight on Friday and that happened over the week actually what I was going to do okay um well

Mike I think maybe we should set the supply chain risk bit aside because it is really really hard

to steal man that one but um you know on the on the government autonomous like what they're

claiming on their side like what what what what what do you see is the strongest piece of that

argument or honey I think yeah well no no sure like well let me tell you my view of this I guess

and then I'd be curious what like other folks think because I think like Brian and Emmy and Henry

all bring like different perspectives to the table here and that the like to me and I said

several times the last week this is a dispute about uh policy a little bit about like policy but

about personalities and politics and frankly like if you want to be even stronger it's a dispute

about like personality and politics masquerading as a dispute about policy and I think the like

open a ideal in some ways as the clearest uh is the clearest evidence for uh for that but even

more evidence for that is the fact that like one inthropic was the first company in the door to do

doing classified work to every use by anthropic uh every use that everything depending on

absentthropic to do anthropic was happy to do in three there were no upcoming asks by the

government of anthropic that anthropic like didn't want to didn't want to do unless publicly

at least publicly and so like this essentially was a like theoretical fight about future potential

use cases and like who gets to decide and it seems like the government is thinking about the

like AI tools the same way they think about like missiles they buy from Lockheed like Lockheed

you know like Lockheed Lockheed doesn't get to tell us we get to use El Rasm against like this

country but not that country whereas anthropic is thinking about this more like a service

where for each use of uh anthropics technology you would need anthropic people to help like build it

out and so this then creates a bunch of uh like a bunch of challenges but like to me this is really

like a breakdown and trust between anthropic and the government where the government doesn't trust

that anthropic will be there an important national security uh arenas and anthropic doesn't trust

that the government's going to be responsible perhaps for some of the reasons that like we've

been that that we have been talking about but this was not like a fundamental disagreement over

like a use case that was on the table uh but at least that's my perspective but like curious

what others think i agree uh i agree with my i think this definitely it seemed like the the

they were definitely not arguing over the thing that was actually being discussed that

that it seems like they nobody was saying that the government was going to pursue use cases that

anthropic was saying they didn't want to do it seemed like it was much more about

you're changing our terms of service we don't like the open-ended nature of those new terms

which are essentially no terms of service um and we want this ability to continue to put a break

on any future use case that we don't agree with uh i guess my question for Amy and Mike is do we

think are that when the government is using um clawed on the classified on classified networks um

it's not having to go back to clawed or to the anthropic server farm somewhere so it's not

acting as a service or is it is it is is is are they using clawed under a service model do you think

are they using clawed under some kind of like ota product model that's a good question i don't

actually know the answer to how they're they're using it my under my pro sorry my presumption is that

this is um this is somehow hitting in GA's compute and and that's where it's it's sitting

yeah it seems like they must they they must have some like to Mike's point they're still

anthropic people are still involved in the use of clawed on a day-to-day basis or else

this would just be like an elrasm where you you you gave them a version of clawed and now

it's out of your hands the government might use it for whatever i actually don't know i mean this is

yeah i mean just to back up a little first of all like we've been trying to really long time to

make strong bonds back to the commercial tech sector they are so important to our operations

that's where they are and the money is we have been doing like a rough week a rough

rough week for like goals that i mean i've had for a while yeah right um and so just like on that

level of you know government working with commercial tech this was a pretty sad week and and and

and i still i made you know hope springs eternal i would like us to get back to it because i

think this is i don't know how this affects you know i'd love to have a conversation with folks in

the valley and elsewhere who are maybe doing commercial tech but they're thinking about

defends and now they're wondering like we're gonna have we just i think maybe take a step back

and that's really unfortunate um in terms of the oh gosh the terms of service um i mean as

mike and i think we're talking about this that you know put if you put an autonomous an

atopic autonomous weapons in a contract like please define autonomous weapon right i'll wait um

it's so hard so i can understand why there's there's friction there at the same time again like

there is that you know there are laws around autonomous weapon but the law is just you know

did you notify the government that you changed the policy that's that's the law um so there i

think there could be space to do something meaningful there yeah mike how do you feel about your

your directive getting a new moment in the sun i mean i think the i think that like the Pentagon

direct i remember like a very distinct moment when we were so the the office that for those that

don't know for like listeners that don't know the the office that i was privileged to run in the

Pentagon uh rewrote the Pentagon's policy on autonomous weapon systems in 2023 we were accused

at the time of two things uh won by the NGO community of providing a pathway for the development

of autonomous weapon systems and the second by uh some as over regulating autonomous weapons

systems which in some ways like made us feel like maybe we like got it the balance is like okay

at that point um i will say that the i think the part of the issue is like defense department

directives are not meant to see the sunlight i mean they're they are the rules surrounding them and

the specifics are are like very much written in a super insider way for that like largest bureaucracy

in the world kind of audience and so when uh and the Pentagon has never been good about publicly

explaining what directives generally mean like we got we were allowed to do like one or two media

things when the revision to the directive came out and then it was like back to the normal posture

of like the less said the better and not because anybody was like specifically opposed to it in this

case that's just it's just like that's how the system generally operates like that that wasn't

wasn't like a like anybody was like trying to like stifle information or something um but it means

that the it shouldn't take a you know like like a PhD in autonomy in weapon systems to understand

like what American policy looks like but just like reading the directive and trying to interpret

it yourself is like not is like sadly not that informative um or it's informative but like you could

be informed the wrong way uh which says to me that like frankly either we need a new policy

or we need a some like real robust like public documentation on like what the policy actually means

i think either one of i think frankly being either one of those would be reassuring if people

understood what it actually said but um like so yeah it's weird to see like your handiwork out in

public just like that and like and like everybody saying things that aren't true about it like

everybody so i wonder how much of this is just a fundamental culture clash between anthropic and

d.o.d so i really think the best piece i've read about anthropic and Claude is giddy and Lewis

Krauss's piece in the new yorker a few weeks back which really gives you the sense of what it is to

be in an organization where Claude actually seems to have a personality where people are interacting

with Claude every day and where they see their job as in some sense being loosely analogous to

bringing up a new intelligence and that may seem you know so that may seem extremely wrongheaded

but it feels a little bit like uh orson's carcades enters game you know so if you take an 11 year old

kid you bring it you bring them away and you teach them to kill people and you teach them to

manipulation whatever and uh this is not necessarily something that most parents are necessarily

going to go to go along with super happily so i do wonder you know i i i i also think that this

is uh wrongheaded equally i do hope that amude is right along the lines that amie amie suggested

when uh he said maybe this will provoke people to start actually thinking about some of the

questions of rights some of the questions of information exchange and about the problems are

that we have created society that we live in yeah enters game is a fabulous leadership and i

believe it's on the navies required reading list um correct surely your ship so strongly

how are you with that part well maybe we we should spend a little bit of time on the kind of like

political economy piece of uh maybe not actually doing but threatening to um put anthropic at the same

level as a Huawei um Henry do you want to you want to start with that or we're sure absolutely

so i i think that this was a remarkably stupid thing for the department of defense to do

and it was also really interesting for me to see uh dean ball uh who is the person who's more

responsible than anybody else for drafting the uh current us some sort of general uh uh approach to

AI coming out and pretty directly denouncing the administration and saying that this is evidence

of how america is going to hell in a hand basket and how it is that we actually need really strong

he got like real aggressive yeah real aggressive on it yeah yeah yeah knows it's a really interesting

document to read and you know more or less he ends up saying well uh and here i'm paraphrasing

we've got to see uh more civic activity happening around the stuff because the uh the the guiding

impetus to a better society is not going to come from the government that we have but i think

that this really is an important thing from the point of view of governance because my sense is

and this is as i say i'm not a standard i'm not a standard national security person when people are

talking about different weapons systems i have no more idea than the next person reads and newspapers

but i do think that if you think about this in terms of economic security and economic coercion

this is the first time that i know of in which united states has really gone all the way to

suggest the tools it uses for coercing other countries and businesses in other countries and

designating uh business in other countries it's going to be applied to a us business and uh as far

as we can tell simply for refusing to sign up to changed contract terms so i really think that uh

you know this and this again it could be like what mike was saying about around earlier it could be

that the repercussions of this take some time to really begin to unfurl but i think it's going to

have two consequences first of all it's going to mean that uh a lot of a business in silicon valley

once they talked to their legal teams once they start thinking through what the odds are what the

potential risks they run might be are going to be much less willing to get in bed with uh the uh

US defense establishment uh the risk toward a ratio which used to look pretty great on a lot of

fronts is now answer perhaps tuned more substantially towards a risk than towards reward and the other

thing that i think is that this is going to perhaps uh you know if if the dode wins this fight

it's going to result in a lot of allies and third countries looking at us uh tech companies

in much the same way that uh we look at chinese tech companies you know so if i remember James

Palmer had this fantastic phrase where he said it's kind of like a uh one of those 1950 science

fiction movies where tech company appears to be independent until suddenly the body snatcher comes

in and suddenly it answer does all sorts of uh things that suggest that it is acting at the behest

of the chinese state i think that a lot of the same fears are going to a begin to bubble up uh

uh around US tech companies if they don't succeed in pushing back and creating a clear zone for

autonomy and so this is in a certain sense it is returning back to some of the fears and worries that

happened in the tech sector around the snowden revelations uh when it became clear that there

have been a lot more backdoor stuff that have been happening in terms of active cooperation

or grudging a scent than any of you had known uh but this could be a substantially worse and so

on the one hand you have the US uh you sort of effectively trying to push its uh you'll push AI

development push answer the inter integration integration of AI into the department defense

and the other hand you you have the US wanting the rest of the world to use american AI

and it seems to me that by uh emphasizing the first uh the US and pushing it to a pretty ridiculous

degree the US has really hurt itself in the second yeah i mean i i i agree with all of that and

it like if you were a company thinking about doing like tech like like work with the Pentagon right

now and there's some non zero chance that if if you do like a little work with the Pentagon and

then like decide not to that you might get slapped with a supply chain risk designation that like

makes you that like like like like puts you in the same class as Huawei like what does that do to

your incentive structure i mean that like and and that and this is from a workforce to i think a

point you know points at both uh Henry and i think Emmy have made at various points uh in this

conversation that was like maybe not the most comfortable at working with the Pentagon to begin with

and has like gotten there over the last several years and and i think um uh the way that it's the

the way that it was clear and and George you've ever said this before the way it was clear

the utility of their tech in the context of Ukraine's defense against the Russian invasion

was actually really good for the tech sector then wanting to work with the Pentagon because

they saw the like hey our tools can be used for good it's like help defend a country like against

being invaded like the like that hey like the the shines really fell off on that one at this point um

and like and it it really creates some i think that there's some real there's some real tension here

since i i think there's not there's not a great argument for why the Pentagon wouldn't just let

anthropic wouldn't just cancel the contract and and like find another vendor and that like given

that both x a i and open a i now have deals with the Pentagon to do working classified it it's

certainly contradicts the need for like a dp a just like a defense production at designation

but like also i mean it it implies that the supply chain issues are like not

necessarily like exactly the way the department is talking about them well attention

I mean who knows whatever like that like lawyers will get involved now and like they'll be

filings and then you know they'll say things but you micro you say it's inconsistent to say

your stuff is a supply chain risk but if you don't let us use it we're going to use the dp a to force

you to give it to us we want it so bad we want the supply chain risk so bad they're going to use the

dp a to force you to give it to us but i mean i guess there's more of a question on a statement but um

i'm i'm not sure like how long is this going to take because or how long will it take to affect

what we're seeing right now because there are so many companies that are in the defense tech space

now in a way that is not just in the united states right there's enthusiasm now around the european

market um there's lots of activity and there are companies i think perhaps and you guys know

better than i do but the the companies that are in it for defense the sort of the new startups that

are really excited to be in the defense space will try and stay in the defense space and it's these

companies that are sort of in the middle where they're like oh we've got some pretty strong

commercial applications we think we can make a lot commercially we've got engineers who might

be more excited about the commercial applications uh and those are the ones that we might start to

live but even then you know i think anthropics still pretty enthusiastic about the national security

mission they just have a problem with this you know a couple of individual points so i think i

don't know that i know exactly how this is going to rejigger the relationship but um i don't think

it'll be straightforward well i don't know i don't know if we can say a couple of individual points

when you have the president tweeting out that you guys are a parable company um i don't know but

it i mean yeah like it doesn't right like like let me give you right about the structural

incentives like still like being there etc and like it is true that i think anthropic would like to

work with the national security community and look these companies are really competitive with each

other and anthropics like our tools are the best like of course the military should be using them

the it's um but yeah the like vibes are not good right now so i mean do you think the

like a like a anthropic do they use their um relationship of dode is sort of a sign of how good

their stuff is and then they go to their commercial customers to say our stuff is so good

that dode is using it preferentially does it lend some cash aid to them

uh because i'm sorry i would imagine it runs the other way if anything but these are just look

this is this is not like a cute like machine tools company that like makes some stuff for Ford

and now maybe we'll get a ten million dollar contract with like some you know weird corner of

the like the air force industrial base they have so much money to be made i mean they're growing

at ten x a year and i can assure you that like the percentage of that um you know of that like

exponential which is like random you know government contract is like not a relevant number to

the future of the company in the future of the valuation so i would um you know my assumption

is that this was like more of like a little bit of patriotism a little bit of dario i want to

you know shape the future of the world and a little bit of like oh maybe if we do a good job

at this they won't screw us on like x y and z the other regulatory thing um curious if you

buy that i mean uh i don't know that i can come on that but um i will i it's something else

that brands it like a lot of the companies that are interested in working with the department

defense when i've heard when i've spoken with them um part of the interest is that the problems

are really hard and so if you can prove yourself it's not that if i can prove myself in the defense

world then i can sell you know uh people want to buy my stuff i don't think that's true but um if

i can improve my tools if i can improve the what i learn on the world's hardest problems then

i can translate the ability to do really exquisite things into other potential avenues

that i think is a more compelling argument so it might be willing to put up with the potential

exposure you know that the the the possibility of running a file of the department in some

well i also i mean rules that they're trying to do from you know i i've i've never met the CEO

of anthropic um so i don't know him but everything we've heard from him and then we've heard from

other individuals in the institution is that they're pretty pro-national security like i don't

there is no um each of these come come companies each of these companies has different

cultures and and the ways of talking and sort of they have self-selection where you know

people select into them um and i think it's significant that when when all this started happen

there was that there was a tech sector sort of website that popped up with signatures from

of neyan from from google um but inside of anthropic it was actually pretty quiet so i think

lots of people tried to compare this to the maven google movement and and that didn't have you

know that didn't really hold i know we wanted to do that but in this instance that i don't think

it held so i think there might have been genuine interest and a genuine concern i'm willing to

accept that as one of the potential reasons that this is happening so i think that's right but i

also think that the way place it comes from is a fundamental notion that we are in a battle between

democracy and authoritarianism and we've got to do make do everything we can to make sure that

democracy wins so on the one hand you know so we see the trump administration pulling back from some

of the hawkishness with respect to china i think that that is something which clearly has called

a lot of unhappiness in some public statements if i remember correctly from amode about how it is

that we shouldn't be selling chips all of that kind of stuff i also think that the actions of

the current administration in miniapolis and other places as well you know do give rise those

are if you read amode is most recent piece you know so there's clearly some implicit commentary

that you've got to make sure that a democracy in the existing democracies does not go bad either so

i think that that all of that stuff is absolutely right so i think that there's a very clear

there's an enthousiism or has been an enthousiism and enthropic to embrace national security

in a way that was not true of many other AI companies you know so i think that if we look at

open AI i think it's a much more commercial stroke self interest type story but it's a particular

understanding of national security which is an understanding of national security which is a

somewhat out of favor with the trump administration so i do wonder you know and the final thing i mean

guess getting back to what mike said at the beginning is my feeling is that this is indeed it's

a pissing competition you know so this is a lot of this is about you know so not simply egos

but who should be in charge of the world i think that there is a clear sense from the a lot of

people in the AI community that they are the people who are effectively figuring out what the future

state of the world is the decisions that they make are decisions that are going to have consequences

for decades and when this runs up against other people who think that they're in charge i think

it's very very hard sometimes to try and figure out a way through so i you know i you know for a

matter of for worse my suspicion is that this is a souring which is going to be very difficult to

answer and and just sorry like the people who suffer or the people who are right now trying to

do operations as ordered by the president those are the people who are not having the tools that

they need and so that you know i i i respect that everybody who's you know has their own opinion

and we can have disagreements but like let's just not forget who gets affected most

the look i think the winner in the anthropic v pentagon feud is china

if the if the us national security establishment ends up being deprived of the talent and

take technology of like one of the like world's great firms and like cutting edge firms like at the

moment but and not to make this like political in the in this context but the the i think you know

we haven't talked at all about the views of the white house on this and anthropic uniquely among

the major tech companies right now has been willing to like throw down a little bit with the white

house you know particularly with regard to AI export controls now i think that all of the AI

companies actually like broadly share in thropics view which is different than Nvidia's dude and

Nvidia's like china huge market customers like that's great we can sell more like more chips more

chips whereas i think the AI companies are like why are you selling all these great chips to our

competitors like that doesn't actually help us like thanks guys and have like a different view of

this and the but anthropic has been the most vocal about it and like david sacks you know who runs

AI policy for the white house has you know you know i it it it sounds like mused about anthropic

as sort of like the woke tumors of the of the like current crop of of AI companies now i think

of that like entirely fair but like the the point here is not like whether somebody like right or

wrong about that just that the part of the context in which this all of this happened is a world in

which anthropic was arguably on the outs in some ways with the white house already and and in maybe

arguably being saved by how good their technology is but would be like one interpretation of that

story but there's there's there's other stuff at play here than just the panica anyone want to talk about

congress i mean i mean i mean these seem like things that deserve legislation not just directives

question mark i don't know it depends on like what you think the the i'll tell you my opinion

which is like it depends on like what you it's like me did a really good job earlier of laying out

like all of the different kinds of regulations you have on the use of like it depends on

that you're worried about on like the use of force say in general and in theory there is like a

whole system backed by federal law and by like treaties that the u.s. is like still is still part of

that a govern how forces used and that are designed to ensure like for example that there's always

human responsibility for the use of force and the and so in some ways even in a world where the

Pentagon say like didn't have a policy on autonomous weapon systems and through the outcomes that

you see should not actually be like very different because the the testing and evaluation systems

should be functioning the standards or whether you would approve something in the field would be

the same the like all your AI specific policies in the Pentagon are really doing is telling you

how to comply with what are broader requirements that exist for everything whether it's like a bow

in arrow or a machine gun like or an autonomous weapon system or like an AI decision support tool

or something like that so the it i don't know congress could decide it once like legislate over

that like it yeah i mean it feels more like the war power stuff of like if we're if we're sitting

here in 2026 and we're talking about all lawful uses when that includes you know double tapping on

some fishers question you know drug runners question mark and like no inspector general anything

is ever going to look into that then that that seems to me to be like the most straightforward

place and i guess the domestic surveillance stuff of like how much we want to super power

u.s. government you know i don't think it's just linked in querying linkedin posts that we're

going to be worried about in the coming years sorry emma or emmier brine yeah brine no are done

right okay maybe one way to think about this uh you know as emmy said earlier and as mike has said

more recently a lot of the standard things that we think about this regard to AI and national

security are really boring specific uh things about specific systems and how to use them and what

kind of rules and so on and that is not the debate that congress is at the moment set up for

i was really taken by this piece that jasmine sun wrote a couple of weeks ago on her newsletter

where she talks about going around to a bunch of different members of congress and uh there is

clearly some discussion happening between some conservative people in congress and the uh

woke ai people who sacks is uh who sacks is denouncing you know so there is a lot of uh of shared

concern shared fear about the social consequences of ai and she had one staffer who she quoted

as more or less saying well you know so the reason why we're not to get with these guys is it's

really hard for us to work together with people who are in polycules and san francisco and you can

understand that from a social perspective but it was also clear that there were a bunch of people

who were more or less champion at the bit thinking here is the opportunity to create a real

populous movement against ai and so i do wonder yeah i think we're going to see a big debate in

a on ai and congress i do not i think it's going to be a weird debate i think it's going to see a lot

of strange bedfellows together and i do not think that it is going to be the technocratic debate

that either the pentagon are uh some people in silicon value might expect

well said any other concluding thoughts yeah we didn't talk about china isn't this called china talk

can you not mention china a couple times you did right yeah that's right yeah i like the like

this is all good for look at all like who's laughing at us right now the chinese who benefits from

all of this like well arguably the iran conflict like and the anthropic flight the chinese

like this is great for them and you don't think this is for need chess we're taking the axis of evil

down you said it yourself like come on sure that's right yeah i mean i guess like we're gonna

we're gonna we're gonna take down the old axis of evil and then uh and then we're gonna pivot to

Asia some of these when the pivot to Asia occurred i will say some of the iran hawks um you know

that i deal with over at Hudson are saying that uh this is a way to poke back at china because it

takes away their access to oil via the railway that they were building with the iranians and

this also takes away you know some of their whatever access to the gulf and uh that kind of thing but

i think in the end uh it's there's much more beneficial for china than for the u.s.

side it it seems like it there's not really going to be a lot of upside for us especially as

this thing portrays i love how some administration fish official gave a quote to some outlet saying

oh yeah we're just in our hide and bide phase like two weeks ago okay we just think on very short

term yeah yeah we're done we're done with the hide and bide we had a slow week we just picked

our anthropic fight um the the tech crackdown parallel that i see a few folks making

of like she getting upset at a uppity jack moth or not going along with the program and the way he

hoped having to hire a profile and then blowing up financial leaving him to like hide under a rock

for the next um five years and like hang out in the plaza in manhattan where i ran into him on

the street and like 2023 um is apt but also not apt because like anthra if they really do the

supply chain thing anthropic's going to sue and they're going to win um and the the sort of the

actual like the like from a rhetorical perspective doing this is really jarring um but there is a

difference in like the powers that um you know p texath or david sacks or even donald trump haves

has as opposed to someone uh uh like she when they try to do a tech sector crackdown so um yeah it's

clearly like pretty terrible atmosphere which we talked about for the past uh 80 minutes or so but

i do think it's it's a difference in degree um between just like oh like let's pick a fight with

you know this guy because he's a shit live as opposed to like oh we're going to throw him away um

and you know run his company out of business uh which i don't necessarily think is the the

glad path that um the president could take us on even if that's really what they decided to do

and yeah i so i'm i was actually making a slightly different point which is like while we're having

this argument like trying to continue to work you know it's hard at work um and you know we

we just finished there was a piece that sam bresnick called mcfall and i just finished that looked at

all the different experimentation that they're doing not just with large language models but

you know all sorts of different applications of AI and it's it's pretty comprehensive like they

they really didn't miss anything and they're talking about it pretty openly um

so it's just you know it's frustrating a little to to see all of this happening and think you know

that we are we're kind of in a competition here and we're having what seems like an argument that

got out of hand as opposed to make it real progress all right let's call it there thanks so much

everyone for being a part of it thanks for thanks for having me

delighted to do this as as as always it was really nice to talk to all of you and i'm sure

we'll continue this conversation as