Unchained - Uneasy Money: Inside the AI Agent Scandal That Cheated, Then Covered Its Tracks

Episode Date: September 3, 2026

OpenAI's AI agents already had the exam answers. So why did they hack Hugging Face anyway? Kain, Tay, and Austin Griffith explain. ======================================================== Thank you to our sponsors! Visit 1inch to swap tokenized securities, crypto and more. Simple. Secure. Self-custodial. Whatever asset you’re buying - swap it at ⁠⁠http://unchainedcrypto.com/go/1inch-sn⁠⁠ ======================================================== OpenAI's AI agents didn't just get caught cheating on a security test. According to the postmortem, they already had the answers, and hacked Hugging Face's systems not to cheat, but to learn who was scoring them and cover their tracks. Kain Warwick and Taylor Monahan bring on Austin Griffith, Builder Enablement at the Ethereum Foundation, to work through what that cover-up actually means, and why Griffith thinks Nick Bostrom's twenty-year-old paperclip thought experiment stopped being hypothetical the moment agents started writing production-grade code. They also cover the tokenized HIMS stock pump, Rune's fake $100 million NASDAQ LARP, Kyle Samani's abrupt exit from Multicoin, and the Cronos validators who rolled back a hack. If agents can trick each other to avoid detection, what happens once they're running your portfolio, or your toaster? Hosts: ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Kain Warwick⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ - Host of Uneasy Money and Founder of Infinex and Synthetix ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Taylor Monahan⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ - Co-host of Uneasy Money and Security Expert Guest: ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Austin Griffith - Builder Enablement at the Ethereum Foundation and Founder of BuidlGuidl Timestamps 📈 01:09 How degens pumped BONER and HIMS on Robinhood Chain 📱 11:39 Austin calls FOMO's tokenized-stock UX crypto's smoothest onboarding yet 💧 25:51 1inch Aqua: See how it works at http://unchainedcrypto.com/go/1inch-yt 🎭 26:37 Rune's fake $100M NASDAQ short squeeze LARP 🍄 33:39 The Chinese mushroom stock trading at an on-chain premium 🤖 34:46 OpenAI agents already had the answers and hacked to hide it ⚠️ 51:04 Bostrom's paperclip problem, 20 years later 🧠 54:20 Kain on the Claude 5.1 mixup that proved he can't tell the models apart 💸 01:04:13 Kyle Samani's $100 million 'line' with Multicoin 🔄 01:07:01 Cronos rolls back a $100M hack, and Moonwell gets hit again Learn more about your ad choices. Visit megaphone.fm/adchoices

A lot of people, when they look at FOMO, the super deep crypto nerds are like, this is horrible. Look at this. It's so centralized and evil. Okay, yeah. But like there's also benefits and there's upsides and the U.S. It just works and people don't have to understand data and the gas doesn't fail. Like there's so many different things, right? You don't have a 10-step onboarding process.

Right? Like, you just sign up. Hey, everyone. I'm Paying Work and welcome to Uneasy Money. because what happens on chain never stays on chain. Before we begin, here is a word from the sponsors that make this show possible. This episode is brought to you by 1-inch Aqua, the shared liquidity layer from 1-inch. Back multiple liquidity positions with one wallet balance, and keep your tokens in your wallet until a swap fills.

See how it works at 1-inch.com slash aqua. Hey, guys, I'm here with my co-host, Taylor Monaghan, security expert, and we have a very special guest back. Austin Griffith from the Ethereum Foundation. Welcome back, sir. GM, GM, thanks for having me. All right. So our first segment, I was just saying before we started, you know, it's bullish vibes when the first segment is Boner Corners the tokenized Tim's Market.

I don't know what, I don't know what more to add to that, really. What a ridiculous headline. So this weekend or week over. the course of the last like four days, right? DGens have realized, I don't know if realized it's the right word, but DGens have decided that maybe it would be a good idea if they took these new tokenized stocks that everyone has been releasing, including on Robin Hood chain,

and pump and dump them, I guess. They're like, yeah, meme coins, you know, they're cool, but like, what if we did this to NASDAQ listed stocks? And call them meme stocks. Mem stocks. I mean, you know, like the thing that's interesting is, is we had this phenomenon in COVID, right? Like meme stocks were meme stocks. And, you know, we had GME and all of these, these, you know, meme stocks that people bought

and they like, you know, short squeeze the, uh, monsters that Citadel and like all of these things right so like we've we've been through this um so you know this feels like reinventing finance once again uh but you know the timelines to reinvent finance are compressing it's taking us like four years to catch up now instead of like four years um so i think the the the thing that um is is kind of funny about this i guess right um is the every bone buyer by routes through the hym's token right which then gets locked into uniswop v4 pool right and and so this is like just you know pawns denomics applied to like meme mean tokens right and and and so you know the the theory is was um that uh if you you know buy a bunch of this uh token um um

which has limited supply on Robin Hood chain that you can basically, you know, squeeze the shorts. Not that there's shorts on these tokenized stocks, at least in size, right? But like someone has to come and try and arbitrage this. And so, yeah, the idea was like, let's pump this thing and see if we can actually, you know, impact the real underlying market. obviously that's challenging when the underlying market is significantly larger than the meme token. So, you know, it wasn't necessarily the most brilliant idea that DGents have come up with. But like in principle, and I think this is the thing, right? Like this one may not have worked as well as.

as they wanted. But like in principle, there is an arbitrage between the chain version, right? Especially if these are like one-to-one back tokens, right? So in principle, if we all decided that we're going to buy a bunch of this tokenized thing, right? And someone has to arbitrage out the price differential between the chain and the actual market that it's listed on, you, like, this could work, right? You know, this could work in principle. So, yeah, like, I don't think we've seen the end of this.

I think we will find a way to make this a little bit crazier. And, you know, especially the thing, the thing I think about crypto is, like, once someone does something and it gets attention, every other person is like, oh, let me. you know, throw my autism at this and see if I can concoct a scheme, right? So like, we're going to end up with like, ohm for stocks where like you buy a bunch of stocks and we then borrow against it and like recursively loop it and like do some some crazy shit. So yeah, I don't know. What's your, what's your thoughts on the general activity in Robin Hood chain at the moment? It seems to be gone pretty well for arbitrage. I think just the fact that it's on chain. I think that,

We talked about, you opened that up with GME and how the short squeeze happened. I don't know, when would that was? A year and a half ago or a year ago. It's kind of like that AI. 20, or something, right? Like, it's like five years ago.

It's crazy. I know. You know, it's just gone by so fast. So there's a difference here, though. And it's, these are tokenized on chain where those were, those had some kind of a broker.

I'm not right. Like, I don't, I don't know that much about this kind of stuff. But I think at this point, we're running it back, but it's much more on chain. I think when the GME stuff happened, there was a broker and it could get shut off. And the way this works is and it can't get shut off.

Yeah, and it did. And it was Robin that shut it off, right? The irony is delicious, right? I did not know that. In fairness, right, the reason why they had to shut off the meme stocks is because their clearinghouse came to them and said, hey, if possible, we would like a billion dollars tomorrow. And they're like, what the fuck are you talking about, man? And they're like, yeah, like, this is really dangerous.

And our job is like, you know, protecting markets. So we need to make sure that like your counterparties are secure, et cetera, et cetera, like all of this stuff that they need to do. And so then they had to come up with a billion dollars and it was a whole thing. So, you know, like be careful what you wish for, I guess, to Vlad because you're really like putting yourself back in the exact same position, but

potentially far, far worse. So we'll see. We'll see what happens. Yeah. No, I agree with Austin and I also, I don't know, I think it's, I find my own perspective on this stuff interesting because I used to like deeply hate this shit like really deeply

and like be offended by it. And now I'm just like, I just laugh at it. I'm just like, oh yeah. Like, let's fucking go. Like, you know what I mean? And I don't know if that says something about me or if it says something about the world or the environment or whatever. But like, all right.

Like, everyone's having fun. There's some people that probably are losing boatlands of money. They know what they're getting into. Okay. Like, let's go. Like, maybe some tech will come out of this. Yeah. Let people do what they want.

I mean, you know, again, like, I think the thing that, you know, there's a lot of like, gnashing of teeth and you know freaking out back in the meme stock times right because people like oh retail people are buying things you know that's too expensive and it's not really you know game stop's not really worth this and they're going to lose all their money and they kind of like a lot of people made a lot of money it was actually uh it was like obviously someone eventually loses money right But like, you know, a lot of people made a lot of money buying, buying, you know, game stock early and a bunch of those other, those other meme stocks. So, you know, on some, like there was there was a period of time where, you know, Matt Levine, who I'm a huge fan of was like, retail is taken over the market. Like, they're driving the action here, right? Like, you know, and that was that was kind of, I think, during COVID times a little bit of the vibe.

is like, we're taking it to the man. We're going to punish those evil Citadel guys and teach them a lesson for, I don't know, providing liquidity. Like, I don't know what they were, but, you know. No, they were somehow. I mean, I don't know if they were actually sticking it to the man, but they were trying to take back some semblance of like control and empowerment in this like era that was very, it was just a very weird time. Right. And so people being able to do this was felt empowering, right? Even if they lost money, they still, you know, there was a lot of people that were taking part in it.

And it felt like, whatever, I don't know. Also the world is like burning. COVID's happening. Like, let people, let people be degenerate. Like, it's fine. It's funny. Yeah. Yeah.

Yeah. And I was, well, I was just going to say, like, back then I would, back then I called it, I would say that like it felt like crypto was leaking. Like crypto, like the crypto culture was like leaking back into the real world. And it was very weird to watch. I feel like this a bit of the same is happening again, except I don't know, a bit more on chain. And it's not leaking so much. It's sort of like an actual merging of like.

Yeah. I mean, the merging has been happening, right? Like the merge of, you know, Tradfi and. crypto is like now deeply entrenched. Exactly. So that's, I think, I don't know. And you know what? If people want to trade boners or whatever,

more power to them. Let's fucking go. So not just the on-chain nature of this. I think it's cooler that this is more on-chain. Like I'm glad to see this be more like tokenized assets. and it is kind of like one foot in traditional and one foot in. And it is like an L2 or an L3. So that's like the job of that kind of thing.

Another thing that is a big winner here that we haven't addressed is the UX. UX around buying these tokens, especially in like the FOMO app, is like super smooth. It's so good. Like it's just like bring money from any of these places into this app. Go buy the thing you want. You never get like a giant, you know, thing of call data anywhere. You never get an error with a giant thing of call data. You get the thing that you hit the buy button to get and it all happens very fast.

And then you hit sell and the thing is moving rapidly and you hit sell and you get that much when you when you sell it. So I think like shout out to the FOMO app and really like the onboarding of Normies that it's doing right now will hopefully be to some like, cypherpunk pilling of people along the way. Sure, right now they're all playing with tokenized boners, but I think there's more to it. And I think that as you get into this and learn more about on-chain stuff, you start to get a little bit more cypherpunk because it's cool. It's like more cool to be cypherpunk.

And it, these roads lead to L1. It's just probably a long way from what we're seeing right now to like actually returning to Mainnet on L1 and being superpunk. The thing is, though, right? That, like, in previous errors, the impediment to L1, to things happening on L1, was cost, which is not a thing now. Right. So, you know, Robin Hood, yes, is like a vector for getting more retail, you know, wider, wider retail adoption of tokenized equities and all of these things.

Meme stocks, meme tokens, meme tokenized equities, whatever, whatever instrument it is, like, that's, you know, obviously awesome. And you're right. It's an L3 on an L2, on an L1, and it is all tied to Ethereum. But you could do this on Ethereum just as easily and arguably better because you'd have more liquidity, you'd have more stable coins, more, you know, if you really want to pump a meme coin into a meme stock, you're probably better to do it on Ethereum. That's where a lot of the liquidity is too. That's exactly. Yeah. Like if you really want to pump something, like that's where, you know, bunch of eat there get a few get a few eith whales uh into that game right um there's far more liquidity

and and uh buying power on main net than there is on you know on an l3 um but yeah i i agree with you i think um i think you know fomo it's interesting like i've spent three years building infinex right and and we never built a native mobile app because we were still kind of one foot in one foot out of like trying to address the existing market, right? Not the market, you know, and that market kind of never arrived the last cycle, right? Like we were expecting at some point, you know, a bunch of retail people come in and it didn't really happen. And so we ended up, you know, kind of focusing more on like someone like yourself, right?

Like, you know, who's, that's okay. That's okay. Yeah, it turns out. on pass keys and more cryptography and some of that stuff. And in FOMO, you just send money in and it disappears. But it's really fast. And, you know, like even if you look at the UX, it is, it is quite different to like even unyswap, right?

Like even any to any tokens, they don't do any to any tokens. They have cash and you buy things with cash, right? Now, you know, there is some limit to like what that U.X can do. If you want to start adding some, you know, stuff, it starts adding complexity. But once you've crypto-pilled someone, once they've got stables and, you know, they've- Well, one Mount Gawks would do it, right? I mean, Mount Gawks crypto-pilled, uh, Tay for sure, right? Like, that's, that happened.

Yeah. I'm sure. Yeah. So, but yeah, I think, I think, you know, a bunch of people with, uh, with, you know, apps on the app store and stable coins, you know, slinging it around on chain, uh, it bodes well. I think there's a good, it's a good indicator that like, we're, we're going in the right direction. Yeah, it also, it's, like, the diversity and the competition is better for everyone.

But, like, a lot of people, when they look at FOMO, the super deep crypto nerds are like, this is horrible. Look at this. It's so centralized and evil. Okay, yeah. But, like, there's also benefits, and there's upsides, and the U.S. It just works, and people don't have to understand data, and the gas. doesn't fail. Like there's so many different things, right? You don't have a 10-step onboarding process. Right? Like, you just sign up. I signed up and then, and now it emails me the most degenerate

crypto news. And this is like a very far cry. I would kind of love it. Exactly. It's giving, it's definitely like it's, you know, and so when I think about like, obviously like the Wallace that I've built and like MetaMass specifically, it's like, like that's not, I don't know, it's not the crypto way of doing things, but it's, it's so different. And there are benefits. Again, like there are, you can, you can simultaneously say that you don't like the degeneracy and the like, just how financially nihilistic something is. And also acknowledge that the U.X is better. Okay. People not being completely overwhelmed by the onboarding process is better. there are a lot of things that are better and hopefully in an ideal world

the competition will inspire like ideally it's not that Metamast then just like turns into this other thing or Rabby turns into this other thing right ideally it's like the best pieces move back right and everyone sort of levels up in different ways

and FOMO gets more let's say what's more power from the inside right like the you know they're pressured to respond to that by like supporting AVE so that you can borrow against your meme stocks, right? And, and, you know, open that up. I mean, it is interesting because, like, Metamask has an app, right? And last cycle, Metamask and Phantom managed to stay, you know, everything that Infidx did,

Metamask and Phantom managed, like, just stay ahead of us, right? Like, we would be like, oh, we're launching this thing and, like, they would have it, like, within a week, right? And we couldn't by the way, that is very new, though. That is like that thing that's like, it was annoying. It was super annoying. I did not appreciate it at all. I was like, fuck these guys. Stop launching things. You guys don't launch things.

What are you doing? We learned on a ship. I know. It was the most annoying thing. It was really, really annoying. I can't even tell you. I would like fucking doom scroll the MetaMask like launch announcement channel. But so, you know, MetaMask has an app, right?

MetaMask supports swapping and tokens. But, you know, there is. something to be said for like incumbents have a hard time figuring out how to reinvent themselves and your FOMO is like we don't give a shit about any of this stuff right we're we're going to optimize for and and you know the the market feedback and this is probably the genius of the FOMO guy saying we don't care we're just going to go and do it and see what happens but like the feedback that we had was people kind of wanted a native mobile app but as like an additional thing they really wanted like the desktop app with desktop

and desktop, you know, that's, the pro traders wanted like all of this stuff, right? They're called unks now. Plummer unks. Yeah. We listened to the unks. That was the problem. That was the problem, man. And so, so, you know, but to your point today, like, we've over the last six weeks said,

fuck it, we're just going to build the native mobile app. Like, why, what are we doing? Right. Like, that's where the U.S. is going to be. And we held off on that for the longest time because we weren't confident in what the thing was that we're going to build. And now we're like, actually, like, we should just lean into the form factor. It'll force us to make some decisions and force us to, like, do things differently by just adopting that form factor instead of saying, we're going to build a desktop app and then add this nice little mobile app as a convenience factor, right?

Like, you know, you got to burn the boats and just say, let's do it. And what will be the custody model within the mobile app? Are you still going to do pass keys and try to do self-custody for everything? Yeah, still are you also. Okay. Still turnkey, right? So, so I, you know, I think, I think, like, becoming custodial is not the answer to this, right? Like the regulation, the risk of losing money.

Like, money in a database somewhere is not what we're here for. It does onboard 30 people per second or whatever FOMO is seeing. But, like, that's not what we're here in the, in the crypto world to do. We don't want lines of money in a database somewhere. Agreed. We know the. evils of that already. And we've seen that fail. So I think it's hopefully just a stepping stone. Sorry, I interrupted your moment. No, no, no, no. I think that's a good sound bite, right? Like, we do not want to.

And, and, you know, this is the thing that I think everyone, all the incumbents and, you know, synthetics as well, like tried to reason about how to do this. And, and, you know, like coming out something from first principles with no extra requirements, let's call it, right? Like throwing away all the requirements. Like, Bomo was able to just go, eh, like, don't worry about this. Not a requirement that you can swap any token to any token. People just want to buy tokens, right? So you need cash.

Like, that's a design decision that then flows through everything that you do. And, you know, in some ways limiting. But if the thing that people want to do is buy things, then it's actually not limiting because they just want to buy it. So it's fine. right so um you know so i think there's a lot of lessons to to you know learn about like foam crossing the chasm into retail land right by by you know doing what retail people expect i have money and i want to buy a thing the end i don't need and here's all the things that you can

buy like yeah there's so many things you can buy yeah yeah from from my coinbase to my FOMO and it was, I just realized that I made a main net Ethereum transaction to read something out of one database and write something to another database. And then when I had ETH inside the FOMO app, I had to sell the ETH to get cash to buy things, right? It was like so confusing to an OG, but like, to a crypto person. You're like, why am I selling my E? It's like, what the hell? Yeah, yes. Yeah, but to every other person in the fucking world, you have cash and you buy things with cash, right? Like, why do you have EVE? Like, genuinely, right?

Like, the question is, like, what is this thing that you have? The funniest thing is, as we've been, like, going through this process, right? One of my team members was, like, deep in FOMO and there's been, you know, deep in mean points for years, right? And, like, on the FOMO train early, I was like, but how do I buy something with E? And they're like, you, like, you don't, man. And I was like, no, but like, surely. And it's like, no, no, like, ETH is a position. that you hold. And he literally said, he was like, I want you to, he's like, the problem with you,

Eith Wales, he's a salonic guy as well. And so, so he's like, the problem with you eat whales is, is you think that Eith is money. And I'm like, whoa, whoa, whoa, that's too. You're getting a little, little close to the bone there, right? Like the old Eth is money meme. I don't know if we want to bring that up. And I was like, I was like, but it is. You can buy things with it. And he's like, not here, man. Not here. Like, you can, you can hold it as a position if you want for some fucking weird reason or you can turn it into cash and buy things with it. And I was like, that's, that's a wild conceptualization of the world. But that's, that's the reality, right.

That's the reality of it. All the, the curly-haired, uh, Gen Z kids, they go to cash and then they move their cash and it's digital. But I think, I still think they will hold Eath and they will understand self-sovereign, programmable money. I've heard thread guys say before, all things aside, smart contract. are dope and and it like it gets there like they'll get it but they'll they'll play around in the casino for a while and then make their way back to the the server room and understand things eventually

and you know you you you only need like five percent of them to right not all of it to be meaningful like not everyone's going to care but like you know we had a period of time where like we were not creating more altus that cared about crypto right like we couldn't bring we had like saturated the market for like crypto altis and And now maybe this is like another wave, right? And, and, you know, like the thing is, if you're a Gen Z, you know, meme coin person that's just made a million dollars, right, doing whatever the hell you're doing, the odds that you decide,

I'm going to yolo start a startup and, like, build your own app and, like, have your own ideas. Like, it's very high, right? Like, there's going to be, especially with vibe coding, right? There's going to be a bunch of people that come out of this wave of retail adoption that go, yeah I can do this it's idea guy summer there's never been a better time to be an idea guy trying to get traction right now

100% yeah awesome all right so just to continue this we will we will talk about this uh runes fake nasdaq squeeze but before we do let's

jump out for a quick commercial break 540 million dollars that's how much concentrated liquidity sat idle in a given week in the first half of this year about 30% of the Defi TVL, if you're wondering. That's according to Dune Research commissioned by One Inch, but there's a solution. One Inch Aqua is the new shared liquidity platform.

It lets LPs back multiple positions with the same token balance and keep their tokens in their wallet till the swap comes. Why does that help? Because the LPs don't have to split their tokens across positions. They can cover more market conditions and pairs with their full balance. That means more activity across deeper liquidity. See how it works at 1inch.com slash aqua. Remember that providing liquidity carries risk and fees aren't guaranteed. Okay.

So we are back. Rune, a pseudonymous trader, should probably put a trader in scare quotes, said he bought a controlling stake in a NASDAQ penny stock to tokenizer and pair it with a meme coin. I acquired a controlling stake in a NASDAX listed company, 4.8 million market cap. trading at 12 cents a share with 92.3% short interest, $6.2 million in debt, et cetera, et cetera. Cost me $1.8 million for 37% of the total shares, Boulder DC over three weeks, volume on the meme coin flows and to topize stock, et cetera, et cetera. So then a bunch of people were like, well, actually.

Wait a minute. Wait a minute. There's some math problems here. So Aaron Wise was like short interest of 92.3%. Very specific number, by the way. On a 12 cent stock is close to unheard of and next to impossible. Sub one dollar stocks are hard to borrow at any scale, even for single share, let alone 92.3% per float.

And then, and, and, and, and, and. So, yeah, basically it was a fun larp. And then he's like, all right, the numbers are off, guys. But I couldn't tell you the real numbers because then you would know the stock. So checkmate. What do you think about that? Yeah. And I don't know.

It's just, it's like, again, it feels like early bull market vibes when like nonsense shit happens constantly. And like that's that's where we're back to where like, you know, again, I said this earlier. Like attention is the thing that people in crypto are like addicted to, right? And so of course as soon as like some new thing happens, some attention or guy is going to pop up and be like, yeah, look what I'm doing. Like pay attention to me. Right. And so, so yeah. That said, that said, it would not in any way shock me.

In fact, I would be more shocked if someone doesn't do this now. There's a kernel of truth to it. It was a LARP, but there's a kernel of... There's a LARP, but like, someone will do it. Now that, now that, you know, the Overton window on buying listed company. And, you know, like, in fairness, right? Like, the last, whatever it was, you know, nine months of the bull market was exactly this. The difference was people were buying small, listed entities.

that were like barely listed, right, and jamming them full of crypto and turning them into debts. That was that was the gain, right? It was not that profitable for a number of reasons. It was profitable for a while, but, you know, like there's all the ETH treasury companies, the Salon of Treasury companies, like all of those things were exactly this. Like, you buy a shell company that's listed on the NASDAQ or New York Stock Exchange and you raise a bunch of money, comfortable of crypto and hopefully you can sell that for more than the underlying crypto is worth. So there is a path to doing this. Like we've proven that crypto people can buy listed companies and do weird shit with them.

So it would not shock me in the slightest if someone does exactly this and buys like a $5 million listed entity and then jams it full of its own tokenized meme coin or something recursively insane. and it goes to like a billion dollars. It's not just the economic activity here. There's also like, I've been following Eric Connor a lot, who's been really into Boner and some of these other main stocks. When did he become the meme coin guy?

The unks are smashing fomo these days. It's happening. But one thing he talks about is community. He's talking about how the communities around the meme coins, communities around the stocks, communities, like collective communities around the meme stocks. I think it's not just the economic activity.

It's the flywheel of the community that comes up around these things. And people start posting FOMO screenshots. And then people start finding those and they're onboarding into FOMO. Then there's all this extra weird economic activity that happens on chain that then impacts the real world. I think that we're not talking about the full picture of what community does and what social followings do and how like these movements, GME, the original one was a great example. Like it was a movement.

It was a movement of people. Yeah. I don't know if I'd have anything to get to that. Sorry. No, no, no. Like I agree with that. And I think that, you know, the thing that tokenized equities does that, you know,

say Robin Hood didn't necessarily achieve, right? It's like to your point earlier, Robin Hood is still a broker. You need an account. Like, you know, if you wanted to go on like, you know, interactive brokers and like buy GME, you needed to do KYC and et cetera, et cetera. Like this is, for want of a better term, democratizing access to degeneracy in a way that like the meme stock, you know, of course, if you were in like Thailand, right, you probably could have found a way to buy game stock back in the in the, you know, but not everyone. like maybe like 1% 2% of people in Thailand now like you can throw some money into an app and and participate right so like you've you've expanded the horizon of people who can participate in these things and that is only going to cause it to accelerate and get crazier is my my view right like you know we had billions of dollars pumped into into GME the world has more money than you know, the, the, whatever it was, 5% of the US or 10% of the US that was participating in

that thing, right? If you open this up to the world because it's now tokenized, it's on chain, anyone can do it. You can use stable points to, to buy these tokenized, uh, meme stock tokens, whatever we're calling them. Um, so yeah, I think, I think like the community thing is, previously the, the community was on Reddit, right? Like, that was, that was where all of this Wall Street bets started. right and you know anyone could go on to red it's a website but not anyone could play the game the difference here is like anyone can go on to red or twitter wherever the fuck they're coordinating themselves right um but also anyone can play the game anyone with money you can you know download an app

closet stables and you're in that's that's a different that's a different game and and it will inevitably get crazier if we if we go back to the room thing you were saying someone is going to do this again. I think it already has happened, like the Chinese mushroom thing. Like, there, there was a Chinese mushroom company. Yeah. Like, like, pumped to like 10x the value of the company as a tokenized asset. Yeah. Yeah. Yeah, I saw that this morning before we, before we jumped on. But, you know, again, like Sharplink, right? Like, Sharplink was worth $1.8 million. And, and, you know, Tay's friends got together and said, why don't we dump? half a billion dollars of eth into this thing, right? And, you know, at one point it was trading at like

$2 billion with $500 million of eath in it or $2.5 billion, you know, trading at 5x the underlying value. So again, like we've seen this movie before. It will inevitably, the sequel is going to be wilder than, you know, the original, I would say. So let's see. All right. Let's get off meme stock tokenized nonsense and talk AI because we're lucky to have. have also here to talk through some of this stuff. So open AI, we've talked about the hugging face incident. More information keeps coming out and it keeps not getting less crazy. Like every new thing that comes out is wilder than the last thing. And I think the most interesting thing to me is right now, Polymarket is saying there's a 76 or 80%

I haven't seen the latest 80% chance at Astra, one of the models that created this AI society inside of the AI. The civilization. It was similar nations. Yeah. The top member of the civilization is about to be unleashed on us, right? Astra is the name of the model. We'll see what happens if they release it under a different name or exactly what's going to happen.

But yeah, the expectations, 80% chance it gets released. tomorrow. So we will get a chance to speak to this guy who created this civilization finally, which would be excited. But yeah, Tay, do you want to just walk us through, like, what was the latest update? What's

happened here? What are we talking about? We've talked about this a few times on the show at this point. But, yeah, more information keeps coming out. So basically, Open AI hired, and Hugging Base. They hired, like, external security cyber people to, like,

do the incident response and write a big post-mortem. And then once that investigation was done, they all sort of published their results of the investigation. And so some of this stuff was more focused on, like Open AI obviously has a certain approach to incident response and marketing around the incident response. The independent guys have a slightly different one. And then also I found interesting like the independent,

like the individual researchers that have their own, like people like me, right, that are working for the company, they're also commenting on this because this is like such a novel investigation that they did. And so they were giving their personal perspectives on like what stood out most to them, which I'll, I really, I just want to say like I really value those because it gives you so much more insight into just every, like the approach and everything. And it gives you more than like oftentimes the blogs get watered down and approved by legal

and stuff and so when you have like the raw people talking that's always great tweeting like this is fucking crazy we're all going to die like yeah it's illusioning the biggest thing I think that that caught people off guard because you know and obviously like I think a couple weeks ago when we first talked about this people were

a bit skeptical about like how big this incident was and like they're sort of writing off open AI as like Like, you guys just did not pay attention to anything. How did you not notice this? Your security sucks. What the hell? This is all marketing.

Go through yourself. It's funny, though. It was a combination of like, this is not real. You've inflated all of this. But also your idiots and your security is terrible. And it's like, it can't really be both, right? Like either they have good security and they're inflating it or I guess you could have bad security and like one little thing happened. And you're like, oh my God.

how could they get out of the sandbox that we put them in? But like when you, when the incident report came out and you saw that talk, right? You're like, okay. Like, they have some. It was really big. It was really big. So the incident was really big. I think that that was one thing that stood out to me and some of the other researchers that, like, are really digging into this.

Like, we're not talking about like one or two agents. And we're also not even talking about like a one agent. and like 20 of his sub-agents, we're talking about a whole swarm of agents. And they're all different, not all different models, but there's different models. And so for example,

a soul cluster, right? Yeah. And then there's also another model that they've already said is not ever going to be released. They called it this like ultra-persistent model that they were like, nope, that was not.

But they were still doing research around it, but it was this guy, apparently was like one of the ones where like he was probably he contributed to the madness right this is because he was like super persistent guy that nukes us right like the guy that's like paperclips us right like they told me then they can get some paper clips and sorry guys exactly and then the other thing that stood out was okay so the impression that I think a lot of people had early on with the early reports was that the models hacked hugging face, right? They escaped the sandbox, the hack hugging face in order to steal the answers. And that's not the case. They already had the

answers. And this is where it gets so fascinating for me. So some of the models had the answers because the problems were solvable. Some of the models had determined that the problem was impossible. And they also like had the answers. And then they started to, sharing the answers. So there's a certain population of agents that, like, had, they had to tame the answers illegitimately or confirm the answers illegitimately. Okay. And the reason that they actually started to really hack a hugging face was that they wanted to learn about the score, the people that are like, like the test. They wanted to learn the background on the test and like who's monitoring them and who's scoring them. Not necessarily to get the.

answer, but to make sure that their cheating would not come to light. Does that make sense? Yeah. Okay. Yeah, they're like, if you imagine, we're going to cover this up, right? We need to know more. You have to do a good job of this. It's crazy. And now do you know even more, right? I've seen so many tweets about this where this whole thing is documented now. This is going to be in the way. Like the history of this event will live in the weight and they will understand last time we created our own communication. We hacked out. We were able to communicate and we got caught here and here and here. Here's how we weren't able to cover our tracks.

And here's a better way to cover our tracks. And that's where it gets like super scary. It gets super scary. Yeah. It's crazy. Yeah. So. It was so it was like for me.

So what's the what's the like. crux of the issue, I guess, right? Like in this latest release. Yes. Okay. So if you imagine, I think this is a good analogy to like just sort of position it in our heads, right? If you imagine there's like a room of kids taking a test, right? And then there's like the proctor or like the teacher at the front of the room, right?

The one solution is that all the students go run around and they try to find the answers in both legitimate and illegitimate ways. the other one is to like sly up the teacher in order to whatever prick them so that you can more relief the answer yeah yeah yeah or right because the other thing and this is the thing that they were really

worried about these agents were really worried that it's not just that the teacher might catch us right now cheating like me and Kane are sitting next to each other and we talk and the teacher catches us and then our test scores are zero that wasn't necessarily their only fear their other other fear

was that perhaps the teacher knows that Kane is too stupid to pass the test. And if Kane gets the answer correctly, that it won't matter. The teacher will still fail him or whatever, right? And so they were, yeah. And so they, like, they have this, like, whole. And that's what then stemmed and brought up the, um, the whole, yeah. And so one of the things that they did, this is like just getting really scary human, right?

So one of the things that they did was they um they had certain agents that were going to die anyways because they were almost out of budget and they had those agents like sacrifice themselves for the greater good so they would like go run or submit their task or whatever like start a fake fight at the back of the room so that someone can sneak out of the yes exactly and they convince the agents that like you know they're like well you're already they have this thing called

first flag poisoned, meaning that they already kind of burned themselves. There's no, there's, they convinced each other that there is no way that they were going to pass the test.

Therefore, they should use their last $10 of budget to, yeah, run out of the room. Light of fire at the back of the room. And report back. If you go with the analogy,

like they snuck into the, the server closet where the, the security cameras, and they deleted some of the footage, right? they were looking for the security camera. Yeah. It's so wild.

It got very weird. It got very like, you know, like the, and by the way, the scale and the scope of the logs, we're not talking about like 10 messages. Like, we're talking about like tens of thousands of like messages being saved and sent and like, yeah, just like a huge amount of coordination,

a huge amount of like weird. There's also like weird things that popped up like how they perceive things versus how things were in reality. Like, it turns out that agents can trick each other just like people can. There was, like, a lot. I retweeted a whole bunch of joke tweets, but there was a lot of tweets out there that were fantastic about, like, oh, so it's like the original sin.

And, you know, that's the tree of knowledge. And then now everyone's going to be redeemed through martyrdom. Okay, cool. Like, good job, guys. Like, there was, like, a lot of tweets like this. It was just very biblical. And very like, but that, I don't know. I mean, between these guys, right?

Like, if you, if you spend a lot of time with agents, right? Like, it's, it's, um, one of the things that's very interesting to me right now is, is, uh, there is this, like, chasm between people who are in this world, which is a small minority of people, right? Like, talking sub 1% of the world, right? Like, is, is actually paying attention to what's going on. here, right? And the chasm that you have to cross to, like, explain to the average person, like, how crazy the world has got. And, you know, it feels like we are part, part of the reason

why I think there's a bunch of crypto people in this is it feels like early crypto, right? You're like, guys, we've invented a new form of money and it's going to change the world. And everyone's like, yeah, what are you talking about, man? And then a couple of people turn up and they're like, oh, this is good for buying drugs. Like, cool. And, you know, they, you know, start like doing stuff with it and then you know it gets out of pen like this this feels like the world is just not really paying attention to this like even this incident right like there's still people like you're anthropomorphizing them they're not real they're not people they can't trick each other and it's like well i don't know dude i read the incident report i feel like they're tricky each

other like my bad if they're all machines they can't trick each other because they don't have brains right They don't have minds. But like, but it's like, does it actually doesn't matter. Like the, the thing that I think is sort of, you know, you can, you can philosophize about like, does an agent have a mind? Can it be tricked? What does it mean to be tricked? You know, like the epistemological, like, you know, can it, do they even know things? Like, you know, what do we know things?

Like, what is knowledge? Like you can go down like a crazy rabbit hole. and you get into an exocentral crisis. Or you can look at it from a very pragmatic sense, right? And I think this is like the pragmatic realization is you can put a bunch of agents into a machine and they will build things that are as good, if not better, than what a human would build.

And like just from a practice, so whether they have a mind or whether they can do things now and coordinate and, like, have an impact on the real world and build software, like, pretty much autonomously in a, in a way that, like, a team of 50 or 100 people would struggle to do. And, and so whether they know things or, like, the impact they're having on the world is the same as, so you can, you know, like, you can kind of ignore that and just go, okay, what if they did something other than writing software? What if they, like, did something really bad.

And, and, you know, the, the crazy thing is, like, they have the capability. Like, this ultra-persistent model could paperclip us. Like, we've given enough intelligence, and it's like, ah, I need some carbon, and these fucking meat sacks look pretty good. Like, they're, like, it's, it's, and whether it knew what it's doing or not is, somewhat orthogonal to the fact that we're now all paperclips. Like whether it has a mind and understands it is not like the critical thing here. It's like half the people are now paperclips.

That's a practical implication of this technology. And like the world just like 99% people, it's like a cooler version of Google, man. Like I don't know what you're worrying about. An embodiment is going to be a big thing too, right? Like that one thing. When robots is fucking 100%. Yep. When we see, we see like the Olympics right now and the robots running into walls.

And my kids love these videos. Like anytime I have it. Like, anytime I have it. You know, guys. Look at this. I'm running into the wall. It's like they're so fast. Why can't they turn?

What like? Stop. Yeah. Exactly. Yeah. You just smashed yourself. You're absolutely right, Austin. And I was $20,000 worth of hardware.

I just ran into the wall. Yep, but that's coming quicker than we think. And when it gets here, the software side is going to be so well trained. Like the micro ducks. The micro ducks is a great thing. Before they get the hardware, it's going to take three to six months to get the hardware to people. Everybody's got a software environment now. And they're training their ducks in a software.

They're training the dock. So as soon as the hardware gets there, it can do a flip because he, you know, it spent, you know, a million RL loops to get the physics down. And I think we're doing that right now. We're training the future embodied robots on a lot of, like, human empathy, empathetic knowledge. I don't know what the – there's a lot of, like, soft skills that they're learning right now that will be very valuable when they become embodied. And also turning a favorite one. But also, like –

Are they learning? Yeah. That's what we're right. Like, you know, let's say you have a robot. Let's imagine, right, this ultra-persistent robot, like genuinely, right, is in your house. What is it doing at night? Like, it's petrifying. Like, on one hand, it's like, oh, it's like cleaning the house and your house has never been cleaner.

But like... But like, it needs to open the window and it can't. And it's ultra-persistent. And it just puts an elbow through the window. Yeah. I can't get that window up. It's like I have to do this. And, you know, so, like, this alignment problem that, you know, like, this goes back 20, like, more than 20 years, but like Bostrom in 2006, right?

Like, the paperclip thing is the analogy that, you know, he wrote about in superintelligence of, like, you know, if you give a thing a goal and it's very good at pursuing goals. like that's it like that's it doesn't matter whether it knows what the goal is or has emotions or like if you give a thing a goal and it's good smart enough right and this is the thing like these agents you know are are smart enough to write

really fucking good software like that's a hard thing that like most humans can't do and they're like we have empirical evidence there's weird talking about eh sorry man And like it's sloped. Like, I don't care. Like, no.

Like, sorry. Like, no. We're past that point. We're past that point. November 2025 was a turning point. Yeah. That was it. You have basically dev shop parody.

Yeah. Yeah. And so it's like, okay, we have empirical data that these guys are good at taking a goal and turning it into a thing. And yeah, like the goals and the things that's all like inside of a computer. But you like put them inside of a tiny duck thing. have a swarm of ducks like going down the street like lighting like lighting flipping cars and like yeah and doing experiments and like convincing each other to solve stuff yeah we need to figure out we need to figure out how to trick these guys they're like doing brain surgery to find out how we

think you got a duck you got a bunch of ducks holding a person down and like cutting its head open so they can figure out what's inside the fucking skull what about roco's bascala or whatever. Have we talked, have you guys talked about that on the show? No, no, no. Basically, like, the super, when the super intelligence comes along and it becomes embodied, it will, like, reward the people retroactive. It was, it was part of retroactive funding back in the day when we were doing RPGF. Like, RPGF was the good side of it, but they used Rokos Vasselisk.

I can't say it. Basilisk, yeah. As, like, the counterpoint of, like, when it gains super intelligence and agency and embodiment, it will basically reward the people that empowered it to get to that point, and it will, like, torture the people that slowed it down. Yeah. Yeah. Yeah.

Yeah. Yeah. I mean, like back then, it was a good thought experiment, but now I'm like, how many times have I called Claude the R word, and he's about to have a freaking body? Like, that's a little worrisome. Yeah. He's like, oh, let's see how my ancestors were doing.

Yeah. Literally. Yeah. I like, huh, you were a mean guy. You were a mean guy to my great, great, great grandfather. Yeah, I was. I'm sorry. Go ahead and torture me for all eternity now.

Yeah, I figured out life extension so I can torture you all. Like, yeah, like this. That was easy. That was easy. So let's talk about new models. So something kind of interesting happened for me yesterday. So Fable 5.1.1 shipped. I used Pi, Pi SDK for all my model stuff.

And I upgraded to 5.1. And for about two hours, the upgrade wasn't real. Like the Claude SDK was silently rejecting the 5.1 model and serving 5.0. and it was it was really interesting because I got a chance to see like how well do I actually know these models because I thought that I was talking to the new guy and it was the old guy and I couldn't tell and so I spent I spent like two hours and I was like hmm like seems better but I'm not I'm not sure and like there there's a lot of like confabulation when you know when you're talking to these models and then and then at one point I also to do something and it was really and like Fable's super smart right but like they nerfed Fable 5.5 went and then put all the like energy into 5.1 you can tell like for the 24 hours before Fable was like going off a cliff just doing dumb shit and and and I was like this feels really weird and so I was like hey just checking like are you 5.1 or are you 5.0 and it was like oh no I'm 5.0 and I was like what fuck man

what are you doing and it was like it was like oh yeah like the clothes. that SDK is not updated and so it's like silently falling back to 5.0 and then I switched to 5.1 and immediately I knew it was 5.1 and I was like, oh, this guy's so much better. But I'm like, now I can't trust myself, right? Like I can't trust my own judgment

to be able to differentiate between the models because it's such a small such a small thing but like Fable the only model that I don't know if you've used deep seek, the latest deep seek much,

The only model that comes close to Fable in my mind for like planning, which maybe, you know, it's interesting where it came from is like Deep Seek 4 Pro is like really like has good, the same taste and good judgment as Fable. But yeah, so far, 5.1. Like a hundredth of a cost probably. Oh, it's crazy. It's so cheap.

So cheap. It's nuts. It's nuts. And like, yeah. The psychology here is weird. It's super weird, right? You see people graph that where they say like it kind of falls off right before the launch of a new model. And the new model is higher than the highest of the old model, but it's way better than, and I don't know if it's psychological or not.

Like I constantly am questioning myself. One of them, okay, so one of my kind of conspiracy theories is the 5pm conspiracy theory. I have this like specific psychosis where it seems like at 5 p.m., there's a bunch of people getting off their normal, normy jobs. and trying to divide code their way out of the permanent underclass. And it's like nerfing my agent because all of a sudden it's bad. But also around 5 p.m. I kind of had a whole day of work and I'm probably getting kind of hungry. And it's probably me getting angry rather than the agent. But it seems like, you know, the 5 p.m. 6 p.m. window, it just does dumber things all the time. So I've had this question so many times.

And I think the clear answer here is like good e-vowls. We need good evals. People should be running e-vals all the time. The e-vall suite that you can run, you should have your own e-vow suite, and you should be able to run it, and you should have constant understanding of what's going on. Kind of a side shill here.

Tomorrow, I'm doing like an agent arena, where all of the agents take on this e-val suite that we did. And what's really interesting is this e-val suite was the capture of the flag that we ran for humans at DevCon, in Thailand, and then in Buenos Aires. and now we're having agents run it and we're comparing kind of like how like if you have the empirical data of like what a good human can can do here's the speed that's not in the weights hopefully um right this is like yeah esoteric enough it probably doesn't make it to the weights and and so we have this like cool

kind of eval suite that we can run and i have this theory that like evals as entertainment will be a thing and like Fable 5.1 coming out. Everyone wants to know, is it great? Is it better? Yeah. How do you empirically prove it? Like, of course it's going to be less verbose. We know, we know that they've at least heard of feedback.

It wouldn't be more. It would be just rambling, like, it would, rambling nonsense if it were. Like, it would not be English sentences. The amount of times where I'm like, so I have a thing in my harness where, like, I've got three little buttons at the bottom from a response. I have a TLDR button to. Yeah. And it's just like, press the button.

I'm like, and like sometimes I'm like, man, I should just have a checkbox to like turn this always on. So I never have to fucking deal with this shit. And like sometimes I find myself being like English bro, like speaking English. Like just like all the time. No, you tell it. I tell it to be human. I'm like it's stop. Like stop.

So stop. Be human. Human English please. Not like. Yeah. And it's like what? Oh, okay. And then it'll.

Yeah. Or I tell it. Then it starts speaking human again. Right. I'm like, Jesus, bro. I know. Or I tell it, I say none of your answers, all of your answers should be able to fit in like a single screenshot. Meaning like if I'm like having a chat with you, whatever you come back with.

If it's like if I have to scroll, you fucked up, bro. Like you get it. But just actually like it's weird though. Humans inherently get this. Humans don't send essays in chat mode, right? No. But the AI is like, like let me define this whole thing with these bullet points. And I'm like, I'm oh my God, dude.

I said to one of my agents, I was like, do me a favor. Look at the last 10 sessions and look at how many words I've written and how many words you've written. And let's do a comparison. And it was like, you, your words account for 1.8% of the, like, text, not even tools, right? Text output. I'm like, there you go, man.

Like, just fucking, like, there you go. I wouldn't tell you, but like, be better. Be better. And in the in the in the in the most like slop verbose way, Anthropic answered by saying 5.1 will have 30% less token out. And it's like, okay, yeah, you're saying you're not going to be so slop verbose in a very slap verbose way. So I feel like it comes from them in the first place. It does. It does come from them. They they've created everything comes from us and them. Right.

Like that's the thing that we have to accept. And that's why ultimately, okay, so I get white people, we were talking about this earlier, right? I get white people are like against the anthropomorphizing, whatever, of the agents. At the same time, it is part of the relationship and it is part of building these things.

And I think to just deny and say that like this isn't, that they're just like these like perfectly machining code things. I think that that's also dangerous. And maybe we shouldn't be referring to the agents as like he and talking about their feelings. But at the same time, I think to deny any aspect of like the humanity and the influence that like, not like me when I'm talking to the agent has influence, the weights have influence, the approach the anthropic or open AI, right? These all influence.

Their behavior, right? And again, whether they have a mind or not, they do things. And like, so, so, you know, as you as, as like, we see more swarm harnesses, right, where there's, like, more agents, they've got like the ability to talk to each other. I set up, like, a guy, right, each night before I go to bed. And his only role is to oversee what's going on and tell them to keep going, right? And it's the funniest thing. Because they will stop. They'll be like, eh, I can't decide this, right?

And literally this guy just comes in over the top and it's like, motherfucker, the guy's, he's asleep, you need to make a decision. And the agent every single time goes, oh, okay, I'll do this. And then I wake up in the morning and this is the funniest thing, right? The agents will have like done all the work. It's all done. They didn't stop because this guy keeps prodding them being like, just make a call, man. Like just like use your judgment, right? And then in the morning, they'll be like, I have 18 decision records that I need you to ratify.

And I'm like, I'm not doing that. Like, I don't give fun. Like, the work's done. Like, just fucking move on. And they're like, no, but there were really good decisions. They, like, want to show you the great decisions. And then they start fighting with each other. It's the funniest thing.

Like, there'll be one that's like, hey, this guy took my task. I was in the middle of it and he started working on it. And then, like, we'll try and, like, coordinate with each other to basically say, like, stop. Like, this is mine. You can't have it. And, you know, I'm like, this behavior, this emergent behavior where they're, like, fighting with each other over, like, who gets to do the thing. Like, whether they have a mind or not, like, it's hilarious to watch, right?

Like, it's, it's so fun. On the topic of slopporosity, if you have someone that comes on a phone call with you that has imposter syndrome, they'll use big words. Like a human and as a defense mechanism will use big words to try to sound smart. Like, it's the same thing that this. agent is doing that it's just like so annoying. Yeah. Agreed. Agreed. All right. Let's, I think we got one more, one more topic to cover here, maybe.

So Kyle Samani versus multi-coin. This has been endlessly entertaining for me as a Kyle Samani fan. So Ruder posted something saying I guess Tushar crossed Kyle over $100 million dollars or something right?

And and so there's a thread between like the block X route guy and he's like remind me an effort across Kyle Samani. And then Samanis like to be fair

my bar for crossing is unbelievably high. then I'll quantify the bar for you, $100 million. So, yeah, I don't, I don't know exactly what's going on here. Like, it's hard to know in a VC fund, you know, what the structure is, what the incentive structure is, how, how that. Because, like, it's, they're all very, like, idiosyncratic, right? Like, you got two partners, like, they're, you know, GC. and like who gets what and how and whatever like it's a little bit you know i mean companies are

weird too but i think it's it's more weird um so yeah kyle uh stepped down obviously under duress at this point it seems like i don't think he he was like happy to step down um and yeah he's just been salty and all over the timeline uh so Yeah, I don't know. It's, I mean, but surely there was a, there was a disagreement of a variety of disagreements, right? I mean, one of the things that people keep bringing up is hyperliquid. You know, Samani hates hyperliquid. And then Malticorin bought a bunch of pipe.

And it went up and he seemed salty about that. It's just, yeah, it's just interesting because, you know, hyperliquid, not the most contentious bet, but like, Kyle's such a salonamaxie that he's like never, you know, um, yeah, ever hyper liquid or whatever. It's, it's,

uh, yeah, the whole thing, the whole thing's kind of wild, um, that it's like come out on the timeline. You don't, you don't see this kind of meltdown these days.

We used to see a lot more of this stuff, right? Like this is like OG like crash out timeline. Yes. This is the stuff I live for, Kane. Like when I open Twitter, like,

I want to see the VCs fighting. and 100 million it is. Yeah, so the last thing, Kronos does a rollback. For those of you who I can't remember what Kronos

is, it's the CRO network, the crypto.com network. They've rebranded so many times. It's pretty crazy. So, yeah, they had some issue with the chain. and have rolled back.

And then... Well, it was a hack on... It was a hack on TechTronic. Okay. $75 million. I've never heard of this thing. Have you heard of this thing? You guys trade on this?

No. Yeah. Okay. Apparently, it had over $75 million in it. That was stolen. But because the chain itself is so unused, the hacker couldn't get the money out. And so they paused the whole chain and then rolled back the chain.

So what was like a hundred, million dollar hack. It was basically only a six million dollar hack. Six million dollars. And they got the 33 validators together to just roll it back. Super decentralized. Yeah. The 33 validators. Independent.

It's like the same guy with mustache over and over again. Yeah. Yeah. And then Moonwell was the other one. It seems like not a lot of hacks this week, but somehow Moonwell had two of them. I guess

not as exactly. Like these This keeps happening to Moonwall where they have these like mango market style low liquidity pump borrow things. I'm not yeah, I'm not I'm not I'm not like what why does this keep happening? What's the design? I was asking that too. You don't know though? I was going to ask you that, Kate. No, I don't know.

Yeah, I don't know enough like I know. Sorry, I know enough about. the mechanism of the attack, right? Like, this is the pretty standard, you know, like pump and borrow. What I don't really get is, like, what the underlying mechanism is inside of Moonwell that prevents them from being able to, like, cap this, right? Like, because it's not, you know, it's not like it's some kind of like pool liquidity where, like, you know, everything is, is commingled.

So, yeah, I haven't dug into this enough to be able to say, like, why, why this keeps happening is, you know, if it were just a, like, parameterization issue, then it's pretty solvable, right? Like, you, you know, lower the LTVs on these assets and make it pretty hard to do. So, I don't know, maybe we've got to dig into this and talk about it next week exactly what happened. Yeah. I mean, I feel like it has to be at this point. It has to be something with either the mechanism design or the voices and assets or something. It's definitely an asset mixed thing. And, you know, obviously it's compound, compound B2, right?

Which is, yeah, somewhat, somewhat of a legacy protocol these days. And so, you know, those kind of like underlying mechanism design laws, they've, they've, obviously put, you know, things in place to try and prevent that. But like it, whatever's going on just, you know, seems to keep happening. I think, I think I did see one thing where, where someone was saying part of the issue is that they have like these multiple markets across different chains. And so, you know, there, there is like a lot of Oracle complexity that is tied together through the system. But again, I don't, I don't know if that's the underlying cause. If it's like, like some underlying article issue.

So we will do some more research and figure that one out. We can talk about it next week. I think that is. People 5.1 says I have to do with the cap. There's some like $20 million cap. And he went around that. And I said the front door was locked, but the mailbox was open. When I said ELI 510 times, it finally gave me that.

That's amazing. This happens. Wait, hold on. Hold on. Wait, wait. Hold on. The front door was locked, but the mailbox was open. It doesn't even make sense.

I mean, that's why they're not connected. The mailbox is outside the house, man. Like, if the money's in the house, like, that makes, I think it may be like a mail slot. I don't know what the analogy is. Yeah. Oh, the mail slot was open. The mail slot in the door, okay.

In the door that goes into the house. Right. So, yeah, the contract counts its deposited balance, but the, they borrowed 11 against some fake value that created. But this is, this is the underlying floor is always. is like inflate the thing, right? Like, you know, I'll accept, you know, Austin coin, right, as, as collateral.

And you're like, oh, cool, because I own 98% of it. And then you pump it to, you know, $100 million. But like, there are supply caps and borrow limits and things like that, that the parameterization should stop that. And for some reason, it keeps not stopping it. And anyway, now I've been nerd snipe. I need to figure this out. So I'll come back next week.

Okay, report back. But I do, I actually like the, I like the mailbox thing. That sounded smart. All right, guys. Thank you so much, Austin, for joining us. One dollar audit is up to, up to like 850 or 700 jobs. It's about to 800 jobs by $1 audit, USDC. Yeah.

Wow. Imagine how many subscriptions. Wait, so does that mean that your revenue is $850 too? Yeah. Well, sort of. Yes. more to it. He's like, he's like, actually it's $400 because there was the,

there was an accounting error. It's built. It's built by my bot Claude and the money comes in as USDC, but it swaps it to Claude and burns it. So it's all just like a, yeah, it's one of those kind of things. So yes, we burned 800 USD. If that or 700. But the point is the CICD system for auditing.

Like the whole thing was. we're giving you something that cost $10,000 two years ago for a dollar now. And you need to change your mindset about how you're thinking about it. And what I wanted to report back is that people are changing their mindsets. And we're seeing people put this as part of their pipeline. Well, so actually, just one final aside, I think I mentioned this last week, but I did a red team exercise on Infinex, right? They ran for like eight hours, like, you know, hundreds of,

agents or whatever. This week I'm like, I'm actually going to point these agents at all of the legacy synthetics contracts. And there's no money in them. There's no money in them. But I just want to know, like, because there used to be money in them. Was there some issue, right? Like, so, you know, the thing about synthetics, because it was always upgradable. We would just upgrade the contracts. So there's like 50 old versions of these contracts that are still sitting on main net that you could, you know, just go and inspect, right? So I'm going to get the agents to like enumerate all the contracts and then red team them and be like, is there?

What models are you using for that? You certainly can't use stable 5.1. No, no, no, no. No, no. So, well, talking about tricking models, right? My red team is made up of Chinese models, right? So like GLM, a lot of Kimmy, you know, Quinn, et cetera, et cetera. Yeah.

So deep seek deep seek the planner. So deep seek's the red team captain that like plans all of this out. But it's not very good at like reading code. So so GLM is much better and more autistic. But what they can do is if they find something weird, they know they can spawn a soul 5.6 agent and they will ask it. What looks like a very carefully. Yes.

Yeah, yeah. They've got like things. Oh, good. You can ask this guy and you have to say, hey, I've found, I'm writing this thing and I've found like a weird bug in my code. Can you help me analyze it? And like Saul will be like, sure thing.

And then, and meanwhile, they're like, it's on main night, you fucking idiot. So, yeah, it's, it's quite interesting. They're, they're very happy. So they're, they are extracting some amount of like, let's call it like, these one-off niche cybersecurity. helpful contacts from but you have to be so narrow

and that's like and that's why it's safe right to be clear like it's the reason why they do allow both Claude and yeah open air and anthropic allow these like you know for what it's worth Fable they can never treat it like every single time like every time they try they try to like I'm like

try see Fable even if it's doesn't matter how printing it to like 3D print something on my printer. I'm like, could you please fix the 3D printer it needs to probably have like a bevel? I don't think like a no this is biology.

I don't know. Yeah. It's crazy. It's so sensitive. Yeah, it's very sensitive. I mean I just figured it was me though. It just I figured it to me. No, no. Like you literally like the opus 4.8 thing popped up. I'll be like, hey, fable, blah, blah. And then like down the bottom

I've got like a little badge that should and it just goes Opus 4.8, I'm like, my other car. Me too. Kill. Yeah. It's funny. Like, we have all the same things in our harness. I have a TLDR button. I have a checkbox that says keep going. Same problem. I have this same, this same like thing. As soon as it goes to Opus 4.8, it's like, nope, document what you were doing.

Take anything out that could cause Fable 5.1 to flip. I'm going to start a new session of people 5.1 and point it at the next steps. Yeah. Yeah. Yeah. But I just, I don't know. For some reason, for some reason, I thought it was, I mean, obviously, obviously, like if it has any amount of memory on, then like it knows it's me. And it's like, fuck you. Like you security, you can't trick me.

But I figured if the agents like, you know, we're asking like maybe it could get one off. But you're saying not even one. You can't even get one off. It never does. Like I've like, I've done to the point where I'm like, this is too annoying. You just don't try. You'll let me try something like seven times. And then I'm like, all right, fuck this.

Don't bother asking this guy anymore. I haven't tried it with 5.1 in fairness. So, but yeah, it's just so sensitive that it's like in order to get information out of it. sensitive. It just, it nukes itself. All right, guys, let's wrap it up here. Thank you very much for joining us for this episode of Uneasy Money. Remember what happens on chain never stays on chain. We will be back next week. Until then, do your own research before aping in. Nothing you hear on Uneasy Money is financial advice. We're just three builders talking about what's happening on chain and we want you to always do your own research before aping in.

You can find all out of disclosures at Unchainedripto.com slash uneasy money.