All right, everybody. Welcome to the literal number one podcast in the world. Three weeks running. We got a lot to get to. Sachs is out this week. We've got a special guest coming up. Let's get to our first story. Ken Griffin says, 'AI is almost certainly a bubble.' And Chamath has responded. Chamath responding and moving markets almost every week here on the pod. Look out, France.
The Citadel founder made the comments this week at the company's annual Future of Global Markets conference, but he says generally, hey, bubbles are normal. Any radical transformation of our economy is generally intertwined with a bubble. He said he's bullish on AI's long-term potential, but his concern is that companies are struggling to turn AI spending into productivity, something we've talked about here, I guess, last year. And so companies are, quote, 'looking at their bills from Anthropic and OpenAI and asking where the productivity is.' We had this discussion over and over again on Thursday. Tech continued to get crushed on the data center and capex front. So people who are either making data centers like CoreWeave, down 10%, or having a huge AI spend, Meta, which is an extraordinary one there, it's down 6% after earnings. AMD down 5%, SpaceX was down 5%. But these things have been swinging up and down a bit and generally trending up.
Chamath, what's your take here on the market sentiment, October 22nd, 2026? What are your thoughts here on how markets are viewing the AI, quote-unquote, bubble? And do you have concerns? Because Jensen from Nvidia thinks money is available and money is going to want this opportunity to finance it. And I think we all think companies are getting a ton of value out of AI technologies.
My Social Capital research team had a really great call with Ramp's chief economist, Ara Kurasian. The first question that they had was because, like, I was doing my own math and I tweeted this thing, Nick, maybe you can put it back up there, where I said, 'Okay, my back-of-the-envelope math, knowing everything that I know and everything that I see, is about 1% of all of the companies represent 90% of all the revenue and revenue for AI spend.' Yeah. So what I said was 1% of companies represent 90% of AI spending. So that's tokens. And of that 90%, 80% of them are from unprofitable VC-backed companies. So the cycle here, just to be very precise, is VCs raise from pension funds and LPs and foundations and family offices. They put money into a startup. That startup takes that money and invests in growth. Part of growing is by consuming other people's AIs. And so that company then starts to show revenue traction. Those folks then say, 'Hey, look how well we're doing.' Raises a bunch of money from the same VCs. And it's just this circularity.
Why that's important is what Ken is pointing out is that for AI to really scale as a predictable driver of long-term earnings, it has to diffuse out of the startup ecosystem. And so you can't be in a situation where on the one hand you're talking about breathtaking valuations, but where 75 cents of every dollar is actually just a startup sending it back and forth so that they can mark themselves up. That's, I think, a structural problem. Now, we've seen that structural problem in the past. That happened in the early days of cloud when AWS initially started to expose EC2 and S3. The consumers of an EC2 instance or an S3 bucket were startups. But the difference there is that AWS was abstracted from Amazon retail, which was already then a roughly break-even profitable business, at least on a free cash flow basis. So what's different here is you have a bunch of money-losing entities just spinning a bunch of money around, and it has to diffuse more broadly. So I think that that's what Ken is observing.
Now, when we brought that data to Ramp, the quote is incredible because this is what my researcher told me: 'I prepared a list of factors to see if your thesis' — this is him texting me — 'around VC money being 80 to 90% of the top 1% of AI spend.' But he said Chamath's actual estimate is pretty much correct. And then the second thing, though, that he said he sent us, and Ramp hasn't published his chart, so thanks to Ramp and thanks to Ara, he said we could be the first ones to publish this.
And just to be clear for the audience, Ramp provides financial services, neobank, credit cards primarily to startups. So they know startup data. There's a free commercial. Yeah.
I think they know broad-based data from other data sources as well, and I think Ara has access to all of it.
But what you said is now look at this chart.
So what this shows is how are the top 1% of spenders in AI spending their money. And I think that Ramp is going to publish this soon, as I said, so thanks to them for letting us preview this. What's amazing to me is the blue box in this histogram. What this shows is what percentage of money they are spending on GPU compute. Now, why does that matter? Well, the rose-colored box is LLM traffic. That's the typical spend you're going to spend on the big frontier labs' closed models, effectively, either through things like Bedrock on AWS or what have you. What's really interesting is that the blue box is ticking up. The blue box is people buying their own GPUs because they're moving to open source. And so what I think is really interesting is you have this two-pronged thing that's happening. One is the diffusion that we need within AI is happening a little bit slower than we would all like.
I see that at 80-90 as well. I think the large companies that we work with, you can almost pinpoint it on a chart. Ever since Alex Karp's commentary on CNBC about data sovereignty, buying patterns have changed, and I think people are being a lot more methodical. They are building out their own data centers or at least moving into VPCs, virtual public clouds. My team yesterday was at the CIO event of, let's say, one of the largest banks in America. You know what they said? I'm not going to say the name of the bank. It was the most incredible thing. They said, 'We have decreased our spend on AWS and GCP. We now have six completely independent, ground-up data centers that we've built from scratch for ourselves.' And I was like, what is going on here?
Wow. Back to the future. People staying in their own data centers. Yeah,
it's back to the future. So, I think if you look at this, I didn't mean to have any commentary other than just observing what I think. But yeah, AI has a diffusion problem. It is too concentrated and ripping around in a small little ecosystem. We need to diffuse out, but as it diffuses out, people are picking open source, is what I would tell you.
And we were early on that discussion two years ago here on the All-In pod. I will say Ken Griffin and I didn't hear the full quotes here because I wasn't in the room where it happened. However, based on the quote, I think he may have something slightly wrong. When startups flood to a technology, that is a strong signal. Why startups — and I invest in a hundred a year and we evaluate over 10,000 a year at my fund — they look for the cheapest, best, fastest edge they can get. What is the biggest edge they could get 20, 15, 20 years ago? It was cloud computing. They saved a lot of money. It was WeWork. They saved a lot of money. It was more efficient. And then it was Facebook ads. Facebook ads were mispriced. They were super cheap. They dumped all their VC money into Facebook ads then. And now it's become AI for the last few years. Yes, they've been dumping into it. So if you see startups rush to a technology or rush to a trend and start spending all their VC money on it, it means this little colony of two to 200-person companies have determined, and they're the most nimble, that this is the best solution. So that's a tell that things are going in the right direction. The big, slow, cumbersome enterprises, they're the slowest to adopt new technology, and that's why they buy startups all the time.
Speaking of buying startups all the time, my guy Nikesh is here from Palo Alto Networks to tell us if this is the end of days and all of our bank accounts, and if Bitcoin is going to get hacked this weekend, and buy more guns and chickens and eggs and drill a well. Welcome back to the program, Nikesh Arora. How are you, sir?
I'm doing great. Thank you for having me, guys.
And you had a nice event last week, didn't you? A little nice event in Napa.
I did. We had a bunch of CIOs, not just one bank CIO like Chamath said. But before we jump into the end of days, I do want to say something to what you guys were talking about earlier about the open source versus the sort of these hyperscalers providing inference. So we had this peculiar problem. We'd like to embrace AI as quickly as many of the startups, and we've gone through all the understanding, all the analysis, open source versus cloud models, and we preference open source, open weight, because it allows us to modify the weights and train them, and something which most people don't talk about is it allows us to interrogate the chain of thought, because over time what's going to happen is all of us enterprises are going to need explainability, auditability, and verifiability.
If I can't validate it, it's not going to be okay to go to your boss to say, 'I don't know what happened, but the IT guys put an AI model in there, it did it.' That's not explainable to the regulator in financial services and healthcare. So they want explainability, verifiability, and auditability, and we can only get that with open source, open weight models. We can interrogate chain of thoughts. That's kind of one reason we're gravitating towards them as opposed to getting caught into the closed race. The challenge, though, I think you guys talked about, and I saw Chamath's chart, the challenge we have is it's very hard to find a global, resilient, highly available inference provider outside of the hyperscale community right now. So if I'm a startup, great. I can go rent 10 megawatts in one data center in one part of the US, and if it goes down for a few hours, I'm out of luck. But I'm a startup, so I'm okay. But I'm providing resilient services at 99.9% reliability. It's very hard for me to find a global inference provider who does not want to sell me their model at this point in time, at the price it makes it economically useful to use open source.
Let me ask you a question, though. Do you think that there's a world where Palo Alto Networks says one of two things? One is you go to the hyperscalers and say, 'Okay, listen guys, you run it for me, but I want the bare metal. I want it totally air-gapped, and it's an AWS facility, but I'll just lease it from you. Otherwise, just get out of the way so that I can have everything inside of it.' Or do you do what this one big GID bank does and say, 'I'm going to build my own inside the United States.' Is there a world where do you think other companies go to that limit, or do you think that's an extreme case that's probably not representative of the norm?
So this is the interesting debate from, you know, six, eight years ago when I came to Palo Alto. I shut down all data centers. I said we cannot run a global data center sort of footprint at the scale, speed, and reliability as I could probably rent it from hyperscalers, and life was great. And suddenly what's happened is in the current sort of incarnation of where we are, compute is highly dear. It's very hard to find compute in the world, period. So I don't — I'm a buyer. I'm a taker of terms, not a dictator of terms. So it's very hard to go and dictate terms anywhere because, yeah, sure, you can make any plans you want, but those plans are, you know, you talk to Jensen, he can't give you chips till 2028. You talk to, you know, where do you get land, power, and shell? So it's very hard to go say, 'Yes, I can go build my own data centers,' but it's kind of an empty threat. It's very hard to go build it. I would be resilient, and by the way, I don't run 50 data centers around the world where I'm running 10 to 20 megawatts of inference. So it's a good conceptual sort of idea that I should go become independent, build my own inference. But most likely the answer is somewhere in the middle. I'm going to have to get bare metal. Yes. Most likely I'm going to get from either a hyperscaler or cousin of a hyperscaler, or somebody's going to be stitching, you know, sleeves of compute from around the world to meet my needs in the hope that I'm going to migrate to something more resilient and more robust over time.
Right. So, you have a hybrid solution down there. One second. I just want to compute thought. You have a hybrid cyber stuff before we run out of time. Just so you know.
Yeah. Then you go and you basically get a bunch of open source models so that you can interrogate chain of thought tokens. Is that a representative solution for most of the Fortune 500 or Global 2000?
I think we get there. I expect at some point in time we will start seeing hyperscalers giving bare metal and letting us own open source, which they already do to some extent. So that's probably what's going to happen because, look, it's a matter of the profit pool, right? It all depends on where you shift the profit pool. Today they all live in Nvidia and the memory industry, and they're being shifted towards the hyperscalers, people who have large compute available, because it's at a point when it will become less about the chips, it will become more about the availability of compute, because I think demand is insatiable. There's enough demand, there's more demand than you and I can imagine. So yes, I understand the enterprise diffusion comment, but it's going to get diffused. The question is, even if you try, if every company in the world today said, 'I'm in, I'm in, I'm ready, I'm embracing AI, let's go as fast as we can,' there's not enough computer in the world to satisfy us, right?
We asked you to come on, Nikesh, not just because you're awesome and you're now a bit based.
He's a bestie. He's a based bestie. He was very quiet under the radar.
I hang out with Nikesh every 48 hours.
I know, David and I keep running into each other.
I mean, every time there's a Knicks-Warriors game, Nash invites me to sit with him, so it's awesome.
That's not true. I'm just manifesting. He's got great seats. I can manifest. A boy can manifest. Kid from Brooklyn's manifesting.
As long as you're willing to go with my 11-year-old, you're welcome anytime.
I'm totally fine with that. I'm sure.
He's always looking. He's always looking for somebody to go with. So, you can be next, Jason.
No, I'm happy to babysit if with the courtside seats or second row.
Exactly. But there's a cyber panic going on. All seriousness, the last couple of days, AI found two major exploits that impact most of the internet. Here's what happened. PHP, which is an open source coding language, powers about 70% of all websites, think WordPress. Wikipedia is at risk. In August, security researchers using AI found two critical exploits nicknamed PHP 2 shell. PHP security team gave this a severity rating of 9.2 out of 10. Basically, Defcon 1 for these security teams. Since August, developers and companies have been working together on fixes. This is standard operating procedure. And the plan was to roll these out gracefully in one major patch on October 22nd, which is the day we're recording this very podcast. In the days leading up to the release, information about the exploits leaked. Nobody knows how it leaked. That's going to be part of the post-mortem of this cyber attack week. So PHP moved the release up just by a day to Wednesday, October 21st. And on Wednesday, all hell broke loose. AWS reported a jump in hacking activity while Cloudflare CEO, friend of the pod, said they saw millions of hacking attempts as well. A bunch of precautionary shutdowns happened. Governments across sites in Europe, France, Germany, Netherlands, Denmark, Finland, etc., they started taking websites down. Wikipedia took down some of its systems just to get them patched. In the United States, the NSA and the FBI put out a bunch of warnings, and now everybody started panicking. People are starting to think that Bitcoin's going to get hacked, their website's going to get hacked, their bank accounts are going to get hacked. It's a bit of chaos. And then there's a second sub-issue around agents accessing banks that dropped at the same time. Now you're in the security business. You've got lots of different products and services here. When did you first learn about this, and how severe is it? Is this an acute issue or is this a harbinger of things to come? And is this going to happen every 10 days for the next year as these models have gotten incredibly powerful at what we call cyber, which really means cybersecurity, not just cyberpunk, but cybersecurity.
Oh Jason, I think 10 days is too long.
I think these things are going to happen much faster than every 10 days. I think part of the challenge we're seeing is we've been talking about technology debt that we all have to go pay sooner than later. And what's happened is people have not paid attention to the idea that they have to go fix their stuff. I had a customer once — but you know, I talked to the CIO this morning. They had a PHP problem. They have a bunch of websites. Some are running the most advanced piece of software. They can put the patch. Some are running older versions which they can't patch. They've got to bring them down and go build them up. But you know what? They never did anything about the fact they were running older versions because it wasn't important. So now they're going to go take those websites down, they're going to go upgrade them, put the patch in, bring them back up. We found out about this a few days ago. We scanned everything in our infrastructure to make sure that there were no issues. We made sure that our products had patches available for our customers who deploy them. So there's two ways to solve this problem, right? One, you can go in and patch the vulnerability. You can go in, find the code, patch what has been given to you by PHP and make it work. The other is you can actually put it in a perimeter defense product like ours, or Cloudflare has one, or other firewalls have one. We put the patch there. It's a virtual patch. So any attacker trying to access that vulnerability can stop at the perimeter because we know what they're looking for.
They're coming to a CDN, content delivery network, first, and that acts as a first line of defense because they're not getting to the actual servers. If they do get to the actual servers, my understanding is they can run code on these servers, and there were some security certificates that were incredibly brittle and easy to hack. The closed models have basically said, 'Hey, these things are cyber risks, and so we need to be very thoughtful.' I know Anthropic did that with Mythos where they brought you in and a bunch of other people and like 50 companies in and said, 'Okay, now go harden your systems.' Were the closed models helpful here? Were they the ones that helped you find these CVEs? Bring us behind the curtain.
The closed models can be helpful because both OpenAI and Anthropic, and most likely Google as well, have very powerful models now which can go — if you give them enough compute, enough persistence, they're going to go find vulnerabilities. They can even chain vulnerabilities to find an attack path, right? If they can solve Navier-Stokes, they can chain for vulnerabilities. And if you give them enough immense compute, they will try every way possible to be able to do it. Actually, the PHP problem was solved by persistent compute. This problem was identified a few years ago, but nobody thought it could be solved because it required too much coordination. So what happened was if you throw compute at the problem, it persisted and it found a way to crack that. Once they found the way to crack that and they published it or got reverse engineered, every hacker knows now how to go after it. Because the challenge you have with patches is hackers are smart. They look at a patch, they see what the patch is trying to do, they reverse engineer from there and say, 'Okay, if I follow the reverse engineer path, if somebody's not patched, I can break in.'
It seems like a bunch of the hullabaloo today that's being drummed up, and this seems like it's going to change every day, is all about these bank accounts getting accessed. Is that, do you think, primarily through kind of the front end, through a website using an agent that's logging in on a browser-based system on behalf of its user? Because it seems like a lot of this may be coming from Muse or some of these other sort of like personal agents. They're logging into your bank account to get information about transactions and what you're doing. And the banks are seeing all this activity and they're like, 'Oh my gosh, is this authorized? Is it not authorized?' The traditional way that banks have provided access to account data has been through APIs using verified applications where I as a user, as a consumer, would log in to my, you know, mint.com account and it had a legal access point to MX or to Bank of America service.
Plaid is an intermediary.
And so there's a bunch of these sort of intermediaries that then provide that capability to apps. Is that ultimately going to lead to a world where the banks are all going to just turn off website access and say no one can log into a website because technically agents could do it and they're not able to kind of manage that? How do you think this is going to evolve as it relates to consumer financial services? Because I actually think this is going to be a big part of the press cycle over the next week or two.
Yeah, I think two different things, David. I think the reason banks have found sort of unauthorized access to their accounts is possibly because of the PHP problem, because their websites are hackable and they've been able to get through them and get to credentials. I think there are two different problems. One is unauthorized access happening because these sites have become hackable. I think the second problem you're mentioning is the personal agent problem. So far most personal agents are using your credentials to log in. So it becomes hard for a bank to decide whether it's actually you logging in or is it your Muse agent who has your credentials logging on your behalf. They cannot distinguish between your traffic or agent traffic. They're both logging in from the website. They're basically logging in as you and your browser and trying to conduct the transaction. Now, as you know, there's this technology called CAPTCHA which tries to see that it's not a bot or an agent accessing the account instead of you. Now, today there are ways to validate CAPTCHA with AI. So you don't even actually have to have the human on the other end. So that's pretty good at it. It's actually sometimes I have to squint to see what the hell it wants me to type.
Yeah. Click all the bicycles in this picture. Like, I fail it three times at our age with our eyesight. But the interesting thing here is there's a hybrid as well, Friedberg, which is your Muse agent, your Grokbot is running a virtual browser. It will pop up the virtual browser and say, 'Hey, do this CAPTCHA,' and then take over. So CAPTCHA is this fear, uncertainty, and doubt, the old FUD by the banks to try to neuter or sandbag these agents from getting the front door to banking, because then they could just route to a bank that's cheaper, better, has better rates on behalf of the agent.
I think that would be giving them a lot of credit.
I think right now the fundamental challenge they have is a liability problem.
Look, it's kind of like if you call customer support at your favorite bank and say, 'I don't know how $50,000 vanished from my account. I never did this.'
Right. And what's the Spearmint Rhino? I don't know what this is.
Keep going. Keep going, Nash. Don't be distracted. Don't be distracted. We have a little fun here. We're number one for a reason.
The challenge is you've given your credentials to an agent you didn't even realize that it got your credentials. It's living on your browser, living on your laptop. It's going to get an innumerable amount of credentials. And I think the biggest problem we're going to have is who owns the liability and who has accountability, whether it's enterprise agents or consumer agents. So I think you see banks saying, 'I do not allow you to interact with me with your agent, period.' Right? They're yelling that disclaimer from the start, saying it's not going to be my fault. Look, banks had to go — don't forget it's not even probably the cost, it's like the liability is secondary. Remember banks, we all know this, like credit cards below a certain amount of money, they don't want to dispute, just pay you because the cost of the customer support call is higher than they're going to have to deal with. Now, if you start having tons and tons of these agents doing it and you start calling your bank for explainability, you're going to create a huge amount of costs. I think that's where they're going to be honest.
Okay. You have a unique seat, Nikesh, because you talk to the people who are being hacked, have been hacked, and you help them prevent being hacked with all these suite of services over at Palo Alto Networks. So I want to take you through a couple of the theories people have about AI's impact on security. You tell me if you agree with them or not. First one is this is going to be largely an issue of who has the most compute, and because the white hats have access to the most amount of compute, they will be able to out-compute the black hats, in other words the hackers. Is that a true statement or is that not yet a true statement, and maybe it will be in the future? In other words, Palo Alto Networks or OpenAI, whoever it happens to be, could fire up so much compute to defend systems that a bunch of hackers in North Korea, Russia, China, etc., just don't have the firepower and they'll be outgunned. True, not true? When will it be true?
True. Not true today because the underlying systems are not architected or orchestrated in such a way that we can launch these, let's call it, defensive swarms, because everything is sort of in dispatched situations, right? Like if it was true, then by definition you could send out a bunch of agents to go patch PHP if you wanted. The company's got 5,000 websites, and guess what, their poor IT guys are going cataloging each one of them because they never actually aggregated everything, put it together, understood it. So I think we need to get our infrastructure to a base level of preparedness, which is where the technological debt lives.
Let me take you to the next thesis and theory people have here, which is persistent computing, reviewing code, and people taking proactive measures more seriously is going to solve this problem. And right now we just have this interim period where people have accumulated technical debt, you've talked about. But AI is going to solve that problem because you'll have persistent tools from companies like yours and others, or in-house ones, that are constantly banging on systems to find these before hackers get to them. Is that going to be the endgame here? True, not true? Short-term, true, long-term, true? Take us through it.
True. Today, if you can see, almost every large company in the enterprise space is out there busy using Mythos or equivalent to try and penetrate their own defenses to see where the holes are, and they're trying to patch them. The challenge, as you'll appreciate, is that the next model is going to be more powerful. It's going to find more ways to get into your infrastructure. So you're going to be playing this cat-and-mouse game of constantly patching. I think there's a slightly different sort of version of this, Jason, where for a second, let's assume that you will get breached, okay? Because there are so many vulnerabilities out there. You can spend a lot of money patching everything. The question becomes less about how I got breached and which vulnerability is going to let people through. The question is how quickly can I find them and how quickly can I eject them, right? So our advice to people is build the infrastructure in such a way, which is possible today, where you're aggregating all the signals, all the data in such a way that you can actually look for anomalous behavior with that white hat compute you talked about, right? Because I have more compute.
Go ahead. And then the third one is the frontier models are causing a lot of this chaos because they are gatekeeping the most powerful solutions to do cyber defense, and that the open source community is giving more functionality to the white hats, to the security teams, to the CISOs to actually solve this, and that frontier models need to change how they deploy this technology in some way. Are they causing part of this problem by gatekeeping this technology and not giving the raw power to the white hats?
What did we say earlier about the diplomatic answer about giving too much credit?
I think the theory is giving too much credit. Again, I think they — remember, majority of them are researchers to start with. They're sitting there and saying, 'Wow, this is extremely powerful. What do I do with it?' And they're talking to us and saying, 'What do I do with it?' Now remember, these models that are so far ahead, they're still trying to figure out how to release them, how to package them, how much compute are these guys going to take, and they're all compute-bound, right? So they're all rate-limiting all of us. Even though I have access to the most recent model which is available publicly, I'm still rate-limited, right? So there's tons of issues that are going on. So I think they're not causing the problem. They're just trying to grapple with the fact that these things are extremely powerful. What they would like, they would like to figure out how to monetize it.
They want to compete with you.
No, they just need money, bro. They're like, they're not listening to your conversations about compute needs of where the money is going to come from. They're like, they're not trying to compete. They're just trying to go after profit pools because they want a lot of money because they got to go give it back to, you know, Jensen.
So that's their main issue.
Let me ask you a hypothetical question. I'll ask all of you. I use one of these agentic services. I'm a consumer. I give it my credentials to my bank account. Let's use your example. $50,000 goes missing. I don't know what happened. Maybe I escalated to the agent I use, you know, Instinct, Muse, or I call BFA. Everybody's finger-pointing at the other one. Maybe there was a prompt injection. Who knows? Maybe the agent just went rogue and transferred the 50K. Let's just assume that that's possible. Or 5K, you know, because it's under some threshold. Who is responsible? Okay, Friedberg. Who is responsible? And who should be responsible? Who is responsible? Who should be responsible?
Well, the terms of service of the bank say you can't use an agent to access the bank account through the website, right?
So, given that that's the terms of service, if the agent hasn't built in a capacity to recognize that and stop that access, then it is a software liability issue with the provider of the agent. I believe unless it alerts you and says this violates the terms of service, do you want to override my block and do you want to do it? Now the liability shifted to you as a user, and we have evidence of this, Chamath, not in banking but in e-commerce. Amazon has blocked these, and subsequently Grok now will tell you, my Amazon shopping bot says I can no longer check out for you, but I can put it into your shopping cart. Tell me which address, which card you want, and I'll tee it all up for you, but I won't hit the buy button. So it's being adjudicated in real time right now. And I think that's probably where we wind up.
I think, come on, we all have to take responsibility for empowering agents and giving them our credentials. You can't say I gave my credit card to Chamath. He went and burned $20,000 on it. Who's responsible? The bank? No way. Is it Chamath? Yeah, he is. But I gave it to him.
Yeah.
Well, what if the agent, Chamath, says, 'I have been programmed to be maximally helpful.' You know, these sycophantic agents, and it goes and says, 'I found a way. You know, I found a way. Actually, if I fire up this type of browser and I simulate human behavior, I think I can get this done.' That's what a lot of these agents will do, is they'll try to please you, and they find that. Then that to me would lean towards the platform providing the agent.
Responsibility. That's you. It's your responsibility.
There's a word in agent. It's called agency. Somebody has to give the agent agency.
Who's giving the agent agency? You or Facebook? I think you're giving agency to act on your behalf. And if you're the one who's giving agency, you have the liability.
Yeah. As a practical matter, it's the only scalable solution because what will happen when these lawsuits happen is everybody's incentive — everybody meaning the corporate incentive — BFA will point to Facebook, Facebook will point to the user. It's the only way that this thing continues to scale like this.
But there's an exception case, and I think Jason said it in his words, you said it, but we didn't dwell on it. It's called the rogue agent. Now, that's a product liability issue. If a product does not behave like it's intended to behave, then there is a platform liability in my mind, which says, wait, you said this is what it does. And today you buy any product in the world, if the product doesn't do what it's supposed to as advertised, it goes rogue on you, then it's the person who built the product, it's their fault. This is where...
territory, which is, hey, it can do 140 tests, it's capable of doing 140 tests, you know, just maybe not really. Let me ask you in the weeds question. We incubated a company, Obliteration AI, and what this company's doing...
Yeah, and Obliterated Models, maybe people don't know about, but they're going to understand what these are. This company, a little bit controversially, provides models that have no constraints, no guardrails, and it's your responsibility when you have a red team, a team that's going out and banging on different vulnerabilities to build up those. What are your thoughts on this new, somewhat controversial...
Wait, before you go, let me just say — before you go, I find Obliteration unbelievably scary. That is a vector of risk that is so unbelievably underappreciated right now because, to your point, Jason, it is a very long-tail thing. Not enough people understand what it is, but I find this probably the thing more than alignment that I think people should really understand and think about. Sorry, Nic, over to you.
Yeah, go ahead, Nash. Tell us what you think of these services.
So I think this is where — this is the dark side of AI, and what I mean by that is, you know, we've had all these debates at the White House. Many of you were present. We talked about responsibility and alignment and making sure we tested them before. The problem is you're dependent on the weakest link. Now, if there's a company which is going to give you the raw power of AI without any guardrails, without any controls that are either built in or can't be built in, then you're basically handing a nuclear sort of bomb to the hands of anybody who wants it because you didn't afford it. That's the problem, because at one level we're saying we need global treaties to manage the pacing of AI or development of AI and just for good. On the other hand, you're saying we can have a company which can allow you to have this most powerful thing with no control, no guardrails, and no governance. It's like saying, 'Hey, here's a DIY aspirin kit.'
Have a good time. You can build your own medication. Have a good time. Sometimes they'll get it wrong, but don't worry about it.
Well, or just giving somebody some really powerful power tools to build things that make things more secure, like build a wall.
I don't understand what the business model of your company is like. Your company is like basically inducing product liability as a service.
No, it's providing this to white hats and corporations. They share the same logo list as Nash and Palo Alto networks. There are people who want access to these in order to do proper testing and make things more rigid. So you do have a higher authority of like not providing this for free like Claude was doing, you know.
Jason, are you telling me they underwrite that these models that when you put these things in infrastructure and do testing, they're not going to go rogue on you?
Well, it's your responsibility to manage the model as opposed to giving it to... Yeah, you if you were the NSA or if you were Morgan, you might need.
How's that different? How is that different from me going and taking open weight model and training it to my needs?
You could do that. This is just a place that has the infrastructure to do it. So providing that infrastructure you talked about earlier like who's going to serve it and that kind of thing and making it more efficient for you to do this with more constraints you know and letting you take the liability but it's a very interesting place.
Why does technical debt build up? And historically, have we just under... and a lot of these causes of these cyber scares? Is this just all because corporations deprioritize security and technical debt and that we just need to change that in this new world and that will essentially solve the problem if we prioritize safety first?
Nobody wants to work on this stuff. I mean, the problem is when you ship something into prod, everybody wants to work on like the next greatest thing. Nobody really thinks about like maintenance as like a sexy idea. That's one thing. Second thing is if you actually deploy it inside of your own virtual private cloud. The problem is the software vendors don't have this one version that's constantly updated. You have a drift problem where you have a version, Nikesh has a version, Freeberg has a version. They're all out of sync. And so now, you as a software vendor are maintaining 19 versions because everybody has their own pace and cadence for upgrades and migrations. Then you have the fact that technologies are constantly changing and the underlying way in which things can happen change so rapidly but the way in which you've done it represents a sunk cost and you can't just rip and replace this stuff and you multiply that by millions of companies and hundreds of trillions of lines of code and this is why it is like it is.
And also you know add to that the incentive alignment problem. The CIOs are not incented to maintain the most modern infrastructure. They maintain to keep cost low and they maintain to keep everything available all the time. So every time you want to upgrade something, you introduce the risk of blowing up or going down. It's like I don't want to go change everything all the time because all my users complain because every time you upgrade something, change something, it causes friction in the system and then it costs money. When's the last time your CIO showed up and you said, 'Yeah, great. Have some more money and have a bigger budget because...'
This is a real tragedy. And you'll see this with like say the CIA or the FBI. If there's no terrorist attacks, people are like, 'We're spending too much money.' And it's like, or we're spending the right amount of money because we haven't had a 911 again. And when they come to us and say, 'Hey, we need more because based on what we're seeing, there could be another, god forbid, 911.' And then Congress or whoever's got the power of the purse at that given point in time says, 'Yeah, we're not going to give you more money because we haven't had one in 20 years.' That is really the core of this problem. It's about to change.
Because we're going to have a bunch of 911s and cyber. Yeah.
Guess what? The phones keep ringing every day and the conversation always is now how do I get out of this because I'm tired of watching get punched in the face every morning when I wake up.
When people really appreciate the CIO and and those groups. When these things the systems come back online and the system's been hacked, it's going to be brutal the next year or so. Yeah. It's going to be a big transition time. Nash.
Well, I think you know the CIOs could become the next CEOs. Think about it. If you believe in three years from now, every business function of yours wants to transform themselves with AI. Let's go to the positive side, right? My CFOs never wanted AI. He's saying, dude, I could do all these anal financial analysis just appoint some people to my team. My marketing people want them. My HR people want them. So, almost every function wants more AI people, which has never happened historically. Everybody wants AI people. If that's what's going to happen, every function is sort of waiting for AI or technology people, they all want the CIO to come and help them. Then who's the person running the company?
What percentage do enterprises spend on security of their total spend and what should it be ideally? I'm just curious. We have no insight here.
The insight is that typically of your IT budget, the cost is somewhere in the 4 to 10% range for cyber. Okay. And 10% is typically where you'll find financial services, the federal government because they want to make sure things are secure and robust. 4% can either be cloud people because that's where the hyperscalers will take care of infrastructure security for you. You just have to worry about application and software security. So it ends up at 4 to 5%. That's the range. You get a 4 to 5% to 8 to 10%.
Yeah. And that's going to have to go up. And there's an issue I think as you're pointing out. If there is AI talent in the building, we got to build new products and services for our customers. YOLO.
Well, Jason, there's a whole... It's kind of like, you know, I was joking the other day, but it's true. Like, look, there's a whole new category which we even don't know how to put our arms around AI. Like, nobody knows like imagine if I said great, you know what, go deploy AI in my company and let's make sure AI is secure. Well, the boss says, whatever you do, make sure you can integrate agents to the chain of thought. That's easy to say. That's hard to do. How do I find all the agents that are running? So it's like we haven't built the AI security harnesses that need. I joke like you know when they invented planes the right brothers are out there. Nobody said let's go create TSA. This is about time we got to build security now because this bad going to happen. So people are busy enthusiastically deploying AI embracing AI. You saw I don't know if you saw this morning that you know with somebody some researcher found that if you send a muse conversation to somebody else you can extract private information out of it. So now they got to go back to the drawing board and patch it. So everybody's experimenting. New shipping being found out. We need to build all the security harnesses for AI. And that's not even started yet.
Bernie Sanders goes on Twitter today and says, 'Look at all the craziness. This is insane. Stop AI now.'
So you're in a room with Bernie Sanders. Someone says, 'Nikesh, we need you to save America. Go meet Bernie Sanders.' You get on the plane. You go over there. You get in the room and the guy says, 'We got to stop AI now.' What is your response? How are you gonna solve this problem? Mahash Aurora, how are you gonna solve it?
I sounds like a great idea. Let's go. Why don't you start? I'm right behind you. It's like, you know, with all without all respect, like...
It's interesting. This is the first time we have dwelled so much on the dark side of technology in history. And I don't know why it is. Maybe you guys can explain to me. We spend so much time obsessing with the dark side of technology. And my thesis is it started with the godfathers or whatever you want to call them the builders of this technology. They've overindexed and dwelled harder on the dark side of this causing this narrative. I think we need a narrative shift. I mean think about it like we only hear about all the bad stuff that this technology does. Oh my god, I got mits. Wait, I got something worse. My guys jumped over. Wait, mine jumped over six months ago. Mine are worse. It's like dude, this is not a good narrative. This is a race to the bottom from a narrative.
We're building a wet lab. But Nikesh, why then do you think Bernie Sanders and others are seizing on the moment? I know we've talked about this before, but like what's going on that's motivating Bernie Sanders to be so vocal about stop AI, stop AI? Is it like a power struggle that they view there to be an emerging system of power that they don't control, that they don't have a hold on that diminishes their power? Or is there something kind of more simple than that?
David, this is where I phone a friend. You guys are much more competent as answering these questions than I am.
Jason and Chimath are probably better equipped to answer these questions.
Yeah. It's a populist issue. If the people at the labs are going to tell you half of white collar jobs are going away and all the, you know, drivers are going away and it's a 10% of everybody dying. I mean, what a great issue to align yourself with if you're a populist, whether you're America first populist on the right or a liberal on the left. It's just a great way to get people's attention and get votes and, you know, maybe it helps you win another election.
Mesh, we appreciate you. You're busy and this is like the busiest time you could possibly have, I think. Right. What is your day like? Have you slept? I mean, are you just getting woken up by clients at like 1:00 a.m. when they're like, 'Hey, I need forward deployed engineers from you.' I mean, what's going on in your life?
I woke up at 3:00. I flew to the East Coast. After this, I'm going to get back on the plane and come back home because I got to make it time for dinner with customers.
Crazy. I mean, it is just a crazy time for you, man. No sleep. I know. Yeah. I love it. I mean, this is the time to grind. All right.
Thanks so much for joining us. We got to keep you on for a full episode and talk about a range of issues. Nash.
You're so follow and follow Nikesh on X. He does a daily. He does one tweet per day and it's based in Nash.
All right. Thank you, Jason.
All right, next up on the docket. Is this the fall of France, gentlemen? We covered this two weeks ago. Situation in France is my god. Well, the place is on fire. It's gotten worse. Over a half a million people have taken part in the protests. They've now spread to Italy, Spain, Albania, and a couple of other regions. Over the weekend, France saw its worst riots in decades. And they love to riot. So, it's a high water mark numbers pretty striking. New York Times data v the French guy. They're good at it. This is their wheelhouse.
Pro pro rioters. Pro riers.
They're professional riers. Exactly. It's kind of their number one pastime. For us, it's like football. For them, it's rioting. Over 3,000 businesses, Chimoth, have been damaged. You've seen the videos obviously trending all over the place. Hundreds of millions in losses. Over a thousand police officers injured. 200 people have now been arrested. 70% of them are adult men. You remember this was a high school protest because high schools were dilapidated and underfunded and that's what the origin of the protest was. Now it's just become complete civil unrest and looting.
Multiple historic government buildings tragically have been burned down. Notably a 143-year-old city hall in a town 250 miles south of Paris. Beautiful name. Beautiful building. Absolutely heartbreaking to watch it collapse. For what reason? So they could upgrade the schools in France. It seems like there's a lot more going on here. France has finally implemented some curfews. I don't know how you don't do that day one. New York Times quote: France's government has accused ultraleft militants and opportunist looters of causing the violence. President Trump and Vice President Vance have called this a third world immigration and welfare state issue. Here's what France's justice minister said. Rioters include quote: people from immigrant backgrounds, people who are foreigners, but there are also people who are fully integrated and who are French. Our partner Polymarket says 18% chance France declares a state of emergency by the end of the year.
Same time investors are dumping French debt. Freedberg. France's tenure has surpassed 5% this week, highest since 2002. On the other hand, and this is an ongoing issue here in the United States, tenure dropped just under 5% for the first time in a month. Wall Street Journal says this might be because people are fleeing EU debt and buying the US debt. Polymarket, our long-term partner here at All-In, says 91% chance Moody's or S&P downgrades France by October 31st. So, let's stop there. Shimoth, when you see this and you spend your time in Italy in the summers, obviously we talk about that here on the pod. You see it happening in Italy and what's your takeaway here from what we're seeing?
It's heartbreaking to see what happens. Let's just be very clear. I think that these people are absolute criminals. They're all terrorists and they should all be arrested. And I think that this shows sort of like what is at the end of all of this quote unquote empathy that all of these governments have been espousing. It's just a disguised way of promoting urban chaos. And I don't know how a police officer thinks it makes any sense for him to be deployed into this melee to do what effectively like protect a Christian Dior store in the seat mall of Paris like it is so stupid. So these folks have created the tinder box and the conditions for this to happen and they've done it because they won't even acknowledge what this is. Even when Macron went on state television, he tried to talk about this in terms of the suffering that these people must have. When you look at these pictures, these are not people that are suffering. These are people that are enacting chaos and violence and terrorism, domestic terrorism. And I think you need to respond to these people with an incredibly heavy hand. And then you need to fix the much broader problem.
And as you said, Jason, I think what's happening now is that France is going to be in an incredibly difficult economic position. We've already talked about this before that France is this critical country that's going to set the tone for the rest of continental Europe in terms of how it finances its long-term debts. It's already proposed 50 billion euros of austerity. That's going to have to go up. I don't understand how in an austerity scheme you actually invest in things like education in a way that would appease these people if that were the true thing that they were rioting and protesting about. The government is in a no-win situation. I think they are going to lose the upcoming presidential elections and I think you're going to see a very complicated reset of the French bureaucracy and the French economy. And I think this is the harbinger of a lot of change to come because the average French citizen should not tolerate what's happening in their country right now.
Freedberg, what's your take here when there might have been completely valid peaceful protests, high school students protesting? It's a beautiful thing. They have some gripe. They created a lot of the democratic system that we actually built America off of. They were great pioneers in democracy and protest. Completely valid form of protest, peaceful, taking to the streets, and then it tips over into something that's obviously at a minimum just incredible criminal and potentially, you know, in Chamat's interpretation, terroristic and the definition of terrorism to terror terrorize people for your political goals. Yeah, seems like it fits the definition. So for you, what do you think this represents? Is this a harbinger of more to come and something we might see flare up here in the United States? Again, remember we had the BLM riots, we had the January 6 riots. We've had our own series of riots here and they had a very similar pattern. You know, people who felt politically like maybe or socially they weren't being represented properly and then in some case it tips over to just criminal looting and incredible violence. What's your take here at Freeberg?
I think it's a little conflating to add some of that other stuff to the mix here.
I think those have very kind of acute political or sociopolitical events tied to them. This one seems to be kind of much more broad-based. So, if you watch some of the interviews that were done on the street, there's a very wide range of kind of complaints. If you listen to the folks that are rioting, everything from schools to free Palestine to the government's not doing enough for us with health care, with support, with jobs. We're underemployed.
There's just so much discontent with the objective being to get the government to do more for the people. There's a lot of immigration that happened. And so all these immigrants that come in, they expect the government to provide to them. The promise was that the government would provide to them. And over time, the government providing things to the people, as we've talked about countless times, is never ever ever ever ever a sustainable answer. The centralized government doesn't have accountability. There is no progress. There is no possibility for economic growth. There is no possibility for creating the right incentive structure. So the more the government tries to provide to the people, the more the people are let down. We've seen this in the United States, I would argue, with the three things that the government is trying to provide here, which is healthcare, housing, and education. And the more the government spends to try and provide more of those things, the worse those things get and the more unattainable they become. And the same has happened in France.
So take a look at this chart. This chart is a measure of public spending as a percent of GDP. Rand Group Research published it and you can see the comparison of France's public spending as a percent of GDP to the peak public spending as a percent of GDP of major communist nations Soviet Union during 1979.
Communist Romania, Venezuela under Chavez.
And then communist China.
That's unbelievable. So do you think France is a communist state?
Well, certainly a socialist state, right?
Yeah, that's my point. And I think this is such an apparent lesson for the West to look at. We all think that there's some political problem with Macron or whoever it was. The more the government spends, the worse the conditions get for its people, not the opposite. And then the people riot and revolt and say that the government's not spending enough and the government needs to be doing more. Now, take a guess. Pull that chart back up. Take a guess where the US is today.
What percentage of our GDP? 20% of...
I was going to say 25. Yeah.
Federal government spending is 7.4 trillion. State and local government spending is 4.6. You add those two together, it's 12 trillion out of a $33 trillion GDP. US public spending is roughly 36% of GDP.
Okay. Yeah, we're right there.
France today is the same as the US. It's about 120% debt to GDP. France today is operating its national government spending deficit is 5.4% of GDP. The US is actually 6.2% of GDP. France is I think probably the first to go because they're only growing their economy at half a percent while the US is growing you know at 2.3%. So we're still kind of hanging on because we have some degree of economic growth that's allowing us to kind of hold things together. But I would argue that we are facing the same sort of fiscal cliff which leads to this extraordinary social unrest as France.
All the economic measures are aligned. We're at the same rate. We're at the same set of problems and we're starting to see the cracks in the system. I think that France is a bellwether for the West. You can load in all the immigration stuff with it because I do think that that's a big part of it where you open the border to let people in to try and grow the economy. You promise them things. They actually add fuel to the fire. Things get worse and it becomes a real problem. Fundamentally, socialism destroys progress. It limits everyone's upside by giving everyone a floor. And the limiting of the upside is what leads to these sorts of problems where no one feels like they can progress. And as soon as the quality collapses because you've taken out all the incentives in the system, everyone starts to feel incredible discontent. France, I think, is one of the first to go. But if we don't right the ship in all these other countries in the west, I think we might find ourselves falling down the same hole. And I'm very concerned about that.
I think we need to observe this for what it is and be really thoughtful about it with our politics becoming unfortunately a little more disturbed going into this midterm and over the next two years. I worry quite a bit.
What do you specifically think should be done?
Cut spending, cut deficit to GDP, cut programs.
Okay, but hold on. Let me give you push back. Yeah. Well, let me just push back for one second. So, you know, what do we think is at the root cause of this? Let's just focus on the young people. The young people, they're not optimistic about the future.
Right. They have a lot of difficulties financially, career oriented as well as social. I mean, like I think their social progression has been stalled relative to other generations. You have these things that are like these swords of Damocles hanging over them like pensions. I'm sure that they don't understand it completely, but you know, if they took a second to realize that at some point these pension things come due for their parents and their grandparents, that's going to be on them. So, how do we set it right for them? What do we do?
I think four things are the primary cause of this that could be directly addressed. It's education, housing, healthcare, and then the retirement system. Every one of them, there's probably a multi-year solution.
Education, I think we need to wind down the federal student loan program. I've said this many times, and I think in the process, we've got to eliminate a lot of the student debt that's outstanding. I think it's a necessity. So, cancel the underwriting at the federal level, but also cancel the existing debt of the folks that have it. There's a complicated mechanism that I think's needed for the deletion of student debt perhaps in bankruptcy or something like that to go back on that law that made it impossible.
Just to add to this, I'm extremely sympathetic to cancelling student debt. I don't think it should be... I think it should have always been dischargeable through bankruptcy, but I'm quite supportive of the...
I think it has to happen with deleting the federal student loan program. What will happen then is you'll rightsize all the institutions and their tuition costs because then all the institutions will start to fit to the market. It'll take about 10 years. It won't happen overnight. So, we have to have a firm commitment to seeing it happen over the next decade. In housing, we've got to do something similar with housing loans that keep inflating the cost of homes and deal with the fact that a good chunk of the middle class...
Okay. Wait, but go through your list. Go through your list because I have one critical question at the end of your list because your list makes sense on the surface.
Then I think that we need to make defined benefit retirement programs, including the Social Security trust fund, illegal. I think all defined benefit programs need to be transitioned to defined contribution programs over call it a decade period. In the process, everyone gets like whatever you're owed in social security gets deposited in effectively an IRA or a Trump account or whatever they're called for you today and all of social security is deleted and now everyone owns equity.
Superannuation funds like we talked about in Australia. We just transfer to that. What would the push back on that be Freeberg? because people lose a guaranteed income. But rather than do it one time, what you would do is you would transition 10% of all your social security amount owed to you per year for the next 10 years. So you do a 10-year transition where everyone gets their money...
