Stable Takes Episode 5: Stablecoin security, global dollars and agent wallets

Stable Takes is a weekly mini-podcast from Tempo on stablecoins, tokenization, and agentic payments. Watch Episode 5 above or read the transcript below.

Tempo's Chief Business Officer, Dan Romero, and Head of Market Development, Simon Taylor, open on the week's multibillion-dollar crypto hack, how stablecoin issuers froze much of it within minutes, and why systems should be designed as if a laptop is already compromised. From there: The Commanding Heights, the essay arguing that AI will drive demand for global dollars, and the GENIUS Act as a foreign policy tool. They close on Link balance, stablecoin balances on card rails and agent spending limits.

This conversation is lightly edited for clarity.

Dan Romero: How's it going?

Simon Taylor: It's going well, my friend. Happy Friday as we record this. One more day inside the singularity. No lack of news.

Dan Romero: Do you guys do Halloween in the UK?

Simon Taylor: Of course. My kids will be getting ready for everything. There's a party tonight, there'll be trick-or-treating, all of that good stuff. We've got spiders everywhere, fake horns, obviously. How about you?

Dan Romero: Same. We have a bunch of costumes, and it should be a good day tomorrow.

Maybe just shifting over: the breaking news overnight is a big hack in crypto. What I saw on Twitter was that it's the biggest hack ever, which is hard to believe given how many hacks have happened over the last decade. But it's scary: $5 billion, give or take.

Simon Taylor: 6.3 was what I saw reported, but it could be 6.5. TRM Labs put out a really interesting study. They've attributed it to North Korean state actors, and they said the vast majority was withdrawn via a mixture of stablecoins and ETH. Tether and Circle have come out and said they've frozen 1.6 billion and 800 million of those funds respectively. So the stablecoin portion seems to have been frozen pretty quickly. But the other assets, like ETH and Bitcoin, were then quickly sold over the counter at 10% above market price, which is really quite interesting. They've been able to follow the money from there, but obviously not necessarily freeze it. So that's the difference between a freezable asset and a non-freezable asset.

The other thing is that Binance has come out and said the attack appears to have started on a developer's laptop. However, they're not attributing fault to that developer. It was actually quite a sophisticated attack that used a weakness on that laptop to move into the infrastructure, and chained a bunch of weaknesses together to get that done. But the scale here is massive. I've seen a few enterprises in the last few hours going, "Hey, am I safe here?" There's a lot of anxiety out there right now.

Dan Romero: I think it's worth breaking this down into a couple of things that are important. The first is that you have all these AI-related vulnerabilities being disclosed, PHP being the most recent. I think something like 60% of the internet runs on PHP.

So the natural reaction is for people to say, "Oh, this must be AI. Nothing is safe. Everything I'm using is susceptible." And there's probably some tail-risk truth to that. These models are getting very sophisticated, and their ability to find vulnerabilities that have just been sitting there for a long time is very much the case.

But I do think a few things stand out, at least a day after, with the information we have. To highlight what you said, one, this was a compromised laptop. So it wasn't a case of "let's break into the server remotely and then just steal the crypto." And from what we know, this was not a vulnerability at the chain or elliptic curve level, which again gets a lot of press. You saw some research Anthropic brought out around being able to figure out the seeds for elliptic curves, which is the underlying cryptography that powers basically all of crypto. I think all of this gets thrown together, and people feel like nothing is safe. But to be really clear: it was a compromised laptop, which in many cases is how a lot of these things happen.

Once you're inside the house, so to speak, with a trusted developer laptop, there's a lot of system access you can get. It's actually something we've thought a lot about at Tempo. Part of this is having folks who had been at Coinbase thinking about it, but with the team we've put together, we spend a lot of time on the assumption that at any given point there is a compromised laptop inside your organization. That's how you should be planning your systems, your defense in depth, and what is available and vulnerable.

I think that has to become the norm in places where traditionally it was a bit more, "Okay, there's the internet, our internet is secure, and if we're inside the house, everything can be trusted." The mentality needs to shift toward the idea that at any given point someone could be popped, so how can you build your systems in a resilient way to minimize unilateral access? There are probably a bunch of details here that we just don't have yet. And I'm not accusing Binance of having the old way of thinking, but I do think it's worth pointing out that you just have to assume at any given point you have a bad actor within your walls, and so building your systems to be more robust is important.

Simon Taylor: I had an old CISO back in my banking days who said we need to change our defensive posture from reducing the chances of compromise to reducing the cost of compromise, and minimizing that as much as possible. I think that's such an elegant frame: something will be compromised, but how do you limit that blast radius as much as possible? And having done that, how do you build in the systems, processes and auditing to make sure it gets dealt with quickly, and that you react incredibly quickly?

But I worry a little bit more about the narrative. To your point, people are seeing smart contract hacks, they're worrying about quantum, they're worrying about AI hacks. There's a narrative issue here that I think concerns enterprises. So being able to talk them through it calmly and say, "Look, here's what's meaningfully different about this situation compared to other situations," is going to be increasingly important.

I'm also given some hope by the fact that both Circle and Tether moved within, I think, 37 minutes to freeze these funds from what I think 95%, 99% of viewers will agree is not a good actor: a state actor in North Korea intentionally compromising consumer funds. I think we want that overall, and I think enterprises want that overall. As an industry, it's a timely reminder that everybody's got to up their game and be constantly paranoid.

I see that inside Tempo too. The paranoia is on a whole other level, and it's great to see. It gives me confidence. As somebody who's worked in banking, I can tell you the Tempo team has some pretty incredible people focused on exactly this stuff. Some of the best I've ever seen in the world.

Dan Romero: Yeah. And I think it's important to also highlight that there are different risks when using crypto, but they can be mitigated if you're thoughtful about system design and you're working with folks who have that experience and can guide you. Not only is it manageable, but in some cases I think it's much more transparent, with the ability to move quickly. You can think of TradFi as a lot more opaque, with a lot less information. The ability for the stablecoin issuers to move quickly here and freeze a good chunk of the funds actually shows me the power of these programmable public blockchains. And if those funds move onto an exchange, someone like Coinbase, Coinbase can take action as well.

So I think there are going to continue to be painful examples, and not just in crypto but across the board. But one thing I think gives crypto an advantage is that every time one of these things happens, everyone is able to level up their posture and coordinate on building the set of controls that enterprises need.

One other thing that's worth highlighting to folks: there's a lot of focus on how fast stablecoins can be, and fast is very much a feature, but the programmability component sometimes gets dismissed, because people think, "Oh, fintech is programmable." There are a lot of system designs where funds move into a smart contract instantly, sit there, and then potentially have some version of a recall period. You can program your own rule set on blockchains.

And I think that's where the industry is going to get to, now that enterprises are interested: built-in compliance controls and, effectively, built-in recall. So instead of having it settle in less than a second, maybe it's okay to have it settle five minutes later with an undo fat-finger button. This is all possible. You can design the system and then enforce the rules in code.

Simon Taylor: The thing that has always surprised me is that we haven't done more with MPC wallets, multi-party computation, for consumers. You can shard a key, or require multiple keys to sign a transaction. I would imagine that if my account was compromised on a centralized exchange, they would be holding multiple keys and looking for risk factors. With my Sardine background, I know there are folks at places like Coinbase who absolutely do this: if it's high risk, they're still going to go, "Are you sure? We want to investigate this first." And I'm a fan of that.

I know that's not for the crypto purist necessarily, but for the crypto purist, have your self-hosted wallet and go move that money that way. You have that option. I think that's not necessarily there for the mass market, and not there for enterprise.

It's not been a slow news week by any stretch. Moving on briefly: have you read this thing called The Commanding Heights? Did you get a chance to read this essay?

Dan Romero: I did. Will had sent it to me. I'm very excited about it, and I know you just did a big tweet today, so I'd be curious about your take before I share mine.

Simon Taylor: Yeah. So the overall piece is, what, 36,000 words, something of that nature? It's long. You've got to sit down with a cup of tea, maybe two, and not get interrupted by your notifications.

I'll do my best to paraphrase the key messages as I saw them. Whenever you try to summarize something that's long, you lose nuance, so I apologize to the authors in advance. But essentially, the US and China will be the two sovereign countries with AI, which creates a quasi-Cold War condition. I think that's fairly well accepted at this point. The durable US advantage will be compute and capital: 70% of compute is now in the US, with 5% in China. The US state will therefore look to own pieces of these companies and build closer relationships with them, as it sees them as a key strategic advantage. Again, I think that's already happened, as we've seen in the press that many of these labs are collaborating directly with the US government.

But this is where it hinges and comes towards stablecoins. China won't be able to monetize its models as easily, because on the global stage it won't let its currency float. That means even the Chinese models will be better monetized by people using the dollar, and the most internet-native, globally available dollar will become the stablecoin, which, whilst not always as tightly controlled as the US dollar in fiat, is globally available in Argentina today.

What this does is create an even more K-shaped economy, where the AI haves and the asset owners continue to get richer and richer, and therefore make up a larger portion of the US economy. Because of that, the US government needs them to keep growing in order to have the GDP to keep funding its balance sheet. So the capex must go up, the US government will backstop that, and it may even limit the yield from US Treasuries to ensure that funding train can continue. Therefore, the US government will look to own them. That starts to displace labor on the global stage, so in markets like India or Asia Pacific, those developing economies start to fall into riotous conditions, and really quite horrific stuff. It's somewhat dystopian as I read it.

I think the core thrust is that the vast majority of the value created by AI will not necessarily go to the labs or the supply chain, as the thesis is today: memory, compute, chips and all that sort of stuff. It will move to the adjacent areas, just as the railroad companies are not massive today, but the economies that were built around the railroads really succeeded.

So there's a lot in here. I'm interested in your thoughts on the stablecoin punch in particular. Of all of the things that go, "Aha, AI will obviously want stablecoins," it's one of the better-reasoned arguments I've seen. I have other questions about it elsewhere, but do you follow that piece?

Dan Romero: Yeah. One, I think it's an amazing achievement to put something like this out there and take a position, whether or not it's right. That's the point of writing the essay, and you can decide.

As a history nerd and a history-of-business nerd, I find the point about railroads very interesting. You naturally have this huge boom period in the 19th century of building railroads, with people like Vanderbilt, all the famous names. But the reality is that the value accrual in the long run, you could argue, went to Standard Oil: Rockefeller and those who could use the cheaper transportation to build new empires.

There's another book, The Master Switch, that talks about how the history of technology usually starts decentralized and then centralizes. Even looking at the last 30 years of consumer internet technology and desktop computing, Apple first got beaten by Microsoft and Intel, only to switch back with mobile. And now there's an argument that it doesn't even matter what the device is if the actual computer is effectively an AI. So I think it's a very dynamic system, and the leading company today may not be the leading company a few years from now.

The one big question in my mind is that if you get into recursive self-improvement, RSI in industry parlance, your model might get so much smarter, and continue to get smarter, because instead of humans building the model it's the AI building the next version of itself, that you may have a durable advantage. TBD. And you just can't magically go create more NVIDIA chips; there are some real constraints in the system. I don't actually know the answer there. I just think it's an interesting question to think about.

But in terms of the stablecoin piece, the thing I think people don't appreciate about the GENIUS Act, if you've spent time in DC, is that a lot of senators and policymakers viewed GENIUS as a foreign policy tool for the US. Take the idea of dollar dominance. We've seen this over the last five years, especially post-COVID: "Oh, the petrodollar is going away." For those who don't follow, you can say global dollar dominance exists because oil is settled in dollars. That's helped make the dollar the reserve currency of the world. Obviously, the military helps with that as well, but there's been a lot of "Oh, BRICS are going to have their own currency."

And now we're in a world where US dollar stablecoins continue to grow. To be clear, Tether, USDC, OUSD, they're all dollars. There are very few non-dollar-denominated stablecoins. With the growth of messaging apps, the primary bank account for you could be an app in your country that is not susceptible to local volatility or inflation, and now you can put it into an earn program and earn what an American consumer can earn in a savings account based on Treasuries. That is a pretty compelling argument to me.

We're actually seeing the early versions of this play out with a lot of the customers we're working with at Tempo. Their initial market for stablecoin payouts plus an earn program and spend is international: a contractor in Argentina who wants to hold dollars and spend dollars. So I do think that's a pretty credible line of thinking.

It then gets into this whole idea that if dollars globally become the place you want to park your wealth, you're actually getting global retail, which for the most part hasn't had as much access to Treasuries, replacing maybe some of the dollars that traditionally were held by sovereigns. There's this whole meme that China has been selling Treasuries and buying gold. For the 7 billion plus people out there who haven't really done this before, it may be better to actually be sitting in dollars that are backed by Treasuries.

And then in a world where you have tokenized stocks, you could even imagine the average person anywhere in the world, with a messaging app that has a stablecoin wallet as their primary financial account, being invested in the S&P 500. It's a bit fantastical, but you can actually think about that, and that's real power for the US to export the financial system globally. There will definitely be pushback, but it's pretty hard. An average small country can't go into an iPhone and restrict the apps. If you're some country in South America or Africa or Southeast Asia, a consumer can have the phone, and that gives them a lot of power, especially with a self-custodial wallet.

Simon Taylor: I think the markets that have the most protection from this are places like Brazil and India, where they have UPI, which is a fast, cheap payments system, and they've been able to use regulation effectively to curb a lot of this sort of stuff. Not all of it: Brazil still has stablecoins, but not nearly to the degree of Argentina.

The IMF did a longitudinal study from 2018 to 2025 of stablecoin adoption in 188 countries, and they found that dollar stablecoin adoption follows currency controls and currency crises almost perfectly. But what's really fascinating is that once the currency crisis was averted and things were normalized in those markets, the dollar stablecoin adoption stayed. So wherever there's volatility, you're seeing these spikes of flight to the dollar and dollarization.

Now, if you're sitting in Europe, that looks like a sovereignty risk, but you also have a fairly strong currency in the euro. And this top-down "we're going to do a digital euro" approach is difficult to distribute in the same way as this bottom-up global dollar, which can be used by any private enterprise. So I can absolutely see that model really taking off. Also, if you're a freelancer in Argentina today and you want to use OpenRouter or ChatGPT, what are you going to use? You're going to use a stablecoin-linked card, because it works immediately and you can hold the value in dollars. So this was probably the simplest, most credible piece of the story to me.

The other part of it is that they say what this enables is the debasement trade continuing at a global scale. I have questions about how far that can really go, and whether you'll see other big developed economies start to push back on that. But you can see the trend here. I think there's something in it for stablecoins for sure. And it's not saying that machines are necessarily going to use stablecoins, which I think is another still fairly woolly concept, but one that has some possibilities to it.

I have more issues with their stuff about calling the 1980s Wall Street guys who invented leveraged buyouts "buyout artists," the dystopian view for some of the long-tail markets, how glorious and magnificent it would be if the US capital markets got to continue, and quoting scripture. Some of this was self-indulgent. But forewarned, you can read this for yourselves and form your own opinions.

Dan Romero: Yeah. And it actually leads us to the last thing we wanted to cover quickly today. With stablecoins, there's a lot of focus on store of value when we're talking about global dollars being held. Unit of account: a dollar is a dollar, that's easy. And then the last core component is medium of exchange, which is where Instinct users now have a balance through Stripe Link, and that balance is stablecoins on Tempo under the hood.

There was a lot of chatter about this on Twitter, and about paying another business on Stripe, and this idea that it's a challenge to the card networks. I think it's worth breaking down, to the point you brought up before, that the stablecoin balance is, at least right now, very much a balance that can be linked to other payment methods. If you look at where consumers are spending stablecoins today, it's a balance that is linked to a debit card. So it is actually running over the Visa and Mastercard rails. And I think people sometimes forget that there are a lot of benefits to those Visa and Mastercard rails for the average consumer, and the ubiquity of acceptance.

Now, if you have a bunch of folks who end up with stablecoin balances, I think it's pretty conceivable that you can start routing some percentage of payments to merchants in stablecoins, depending on the use case. That's what I'm excited about in the long term. But I think people should almost pump the brakes a bit. The sequencing here is to get people to have stablecoin balances. If the balances are there, you can actually spend them. Then you get people comfortable that they can spend from this balance, and once people are used to spending from the balance, you can potentially optimize certain flows to say, "This is a stablecoin-to-stablecoin flow." I'd be curious about your take.

Simon Taylor: What I love about this product is that you don't see a stablecoin, you just see a dollar balance, and a dollar is a dollar. I think that's the future.

The other thing is that there's a real, material user benefit. If you're using Instinct today and Muse, it stops quite often and goes, "Can you just authorize this payment for me?" Maybe it's a step-up authentication, maybe you have to click something on a website; it's still breaking. A lot of that's coming from merchants, and there are a lot of good reasons for it.

With a balance, you don't. It's essentially like giving your corporate card to an employee and saying, "You have this limit, and you can just go spend stuff." Of course, that stablecoin balance is then attached to a prepaid card, which actually completes the purchase, but by creating this balance, you're in control of it, and it can spend up to that limit. I think it's just a really elegant, low-friction user experience to say, "Agent, you have $700 this week. Manage all of these expenses, go do it, and just keep me updated." That feels like the future.

And again, it feels like the kind of thing where a stablecoin is just lighter weight. Could you do this on an existing banking-as-a-service FBO account structure? Yes. But there are some nice advantages to the lightweight nature of a stablecoin: it doesn't require additional overheads, it's easy, it's programmable, it's simple, and I can attach it to the card infrastructure later. So simplicity and cost are actually important dynamics here.

Dan Romero: Yeah. The thing I'd finish on is that if you go back to the Satoshi white paper, this is actual digital cash that is programmable, and that makes it really easy, even if you're routing it over card network rails: the ability to spin this up, put the permissions on it, and natively enforce that at a core primitive level of the wallet and the chain, whether it's a smart contract or not.

And to your point, you can do this with fintech today. That's the classic thing a stablecoin-skeptical fintech person says: "Oh, well, I could just go do this." And it's like, "Well, give me the product. I'd love to go try it." And they don't exist. Money stuck in a bank account that you then need to move to an intermediary service to go do that is not nearly as effective as my primary balance actually being stored in stablecoins. There's a financial or economic incentive for me to go do that, and then I get all the benefits of the programmability, and of being able to just have my agent get the job done for me without any of the friction.

Simon Taylor: Well, I'm for it, genuinely. A lot of people were responding, when I covered the news on my social accounts, with "Well, what's the benefit here?" And I'm like, cost, time, simplicity for developers. I think sometimes people look at a new technology and want some sort of fourth dimension to emerge in which they discover some new law of physics. And I'm like, well, no, the laws of physics are still very much the same. Better, faster, cheaper is really, really good, and programmable is really good.

Let's come back and do this again another time. Good to see you.

Dan Romero: Good to see you.