The Information's TITV
00:07:48 · Yueqi Yang, Akash Pasricha
[00:07:48] S2: Hey, Akash. Thanks for having me. [00:07:50] S0: So, I want to start with the risks here. When you've been talking to your banking sources about these agents, I mean, on one hand, there's using agents in your own course of work, as an employee at a bank. On the other hand, there's customers using agents for some of their banking products. What are some of the risks here on both sides here that have been coming up in your conversations? [00:08:20] S2: So I think bankers are generally pretty concerned and cautious about people giving their logging and bank account information to their agents, especially as agents are becoming more popular, like Meta's Muse and Instinct are taking off very quickly and people wanted to use these agents to do shopping for them and potentially looking into their personal finances. And we've heard from the big banks saying that their policy is to block the agents. And that's because they're worried that agents might go rogue. And if your agent is misspending your money, then it's unclear who is going to be liable for that money. But at the same time, I've also talked to some of these more FinTech friendly banks, such as Lili Bank, Cross River Bank, Column, and they're actually going in the opposite direction. And for them, agents is just another extension of what they've provided before, which is to provide API access for banking use cases. And they're the early movers that are embracing agents to allow them to access bank accounts by consumers.
00:08:20 · Yueqi Yang, Akash Pasricha
[00:08:20] S2: So I think bankers are generally pretty concerned and cautious about people giving their logging and bank account information to their agents, especially as agents are becoming more popular, like Meta's Muse and Instinct are taking off very quickly and people wanted to use these agents to do shopping for them and potentially looking into their personal finances. And we've heard from the big banks saying that their policy is to block the agents. And that's because they're worried that agents might go rogue. And if your agent is misspending your money, then it's unclear who is going to be liable for that money. But at the same time, I've also talked to some of these more FinTech friendly banks, such as Lili Bank, Cross River Bank, Column, and they're actually going in the opposite direction. And for them, agents is just another extension of what they've provided before, which is to provide API access for banking use cases. And they're the early movers that are embracing agents to allow them to access bank accounts by consumers. [00:09:40] S0: So, we'll get to the smaller banks here in a second. But the big banks, I mean, have they said outright, we're not going to allow agents to access accounts or anything like that?
00:09:40 · Akash Pasricha, Yueqi Yang
[00:09:40] S0: So, we'll get to the smaller banks here in a second. But the big banks, I mean, have they said outright, we're not going to allow agents to access accounts or anything like that? [00:09:50] S2: Yes. So, JPM, Jamie Dimon specifically said during their earnings call recently that they're going to block all the third party agents. And for some of these other banks, Wells Fargo, Bank of America, for example, recently, they've changed their terms to say that if a customer is giving their login access to their agents, and then the agents get hold of their account, then the customers are responsible for any actions taken by the agents, presumably including agents misspending your money. So that's the initial steps that these major banks have taken to block or at least try to block external agents. [00:10:37] S0: Okay. Now, about the smaller banks, Lead Bank, Cross River Bank? What's their approach here?
00:10:37 · Akash Pasricha, Yueqi Yang
[00:10:37] S0: Okay. Now, about the smaller banks, Lead Bank, Cross River Bank? What's their approach here? [00:10:43] S2: So the smaller banks are working with FinTech providers to provide access to AI agents to use bank accounts or to use cards that are prepaid and with spending limits. So for example, Lead Bank works with Stripe and Stripe is one of the most important and dominant payment provider for a lot of popular AI agents, including Manus and Instinct. And Stripe is able to provide a single use virtual cards for agents with a spending limit so that they can go shopping for you. And these cards are issued by banking providers and Lead Bank is one of the providers. And some of these other banks, Column, for example, they work with a startup called Natural, which provides digital wallets for AI agents. And these wallets are essentially subledgered accounts under the name of the human users held at open at column. And that effectively means that the money held in the wallet for agents are actually FDIC insured money and account at the bank. [00:12:04] S0: So, if I'm hearing you correctly, I mean, it sounds like the smaller banks, they're not necessarily saying that the agents are not going to go rogue. They're just saying that if that happens, we're going to play in a much smaller sandbox of capital. Is that the idea?
00:12:04 · Akash Pasricha, Yueqi Yang
[00:12:04] S0: So, if I'm hearing you correctly, I mean, it sounds like the smaller banks, they're not necessarily saying that the agents are not going to go rogue. They're just saying that if that happens, we're going to play in a much smaller sandbox of capital. Is that the idea? [00:12:23] S2: That's exactly right. Like for a lot of these banks, they have spending limit, they have transaction limit imposed on either the card or the account for increased bank. For example, I talked to the owner of increased bank and he said the bank is planning to offer isolated bank accounts for agents, which means that the agent won't be able to touch the rest of the money owned by the human customer. So the intent for all of these setup is to isolate the exposure that the agent can create if they go rogue or they misspend your money so that the responsibility and the liability is limited. [00:13:08] S0: Doesn't exactly scream trust, in my opinion, of these tools, if you ask me.
00:13:08 · Akash Pasricha, Yueqi Yang, Paddy Srinivasan
[00:13:08] S0: Doesn't exactly scream trust, in my opinion, of these tools, if you ask me. [00:13:15] S2: Yes, but I think that the argument that these banks and some of these fintech startups make is that in the future, you will not be able to block agents from accessing your bank accounts in part because agents also have access to your email. So if you think about how you log into a bank account, you generally get a verification code that's sent to your email or your phone, and the agents can use that code too. So they think rather than just trying to block it, which you might actually fail, you should provide a controlled scoped access for agents so that they can at least be able to tap into your isolated separate bank account, do what they need to do, but then you have an auditable rail to see and track what exactly your agent was doing. [00:14:07] S0: Now, Yueqi, this is interesting because between the big banks and then the small banks, I mean, I don't want to call them a middle player here, but a separate player are the trading companies or trading brokerages like Robinhood, and we had the CFO on the show a couple days ago. We were talking about their AI agents that they're using. I mean, these trading brokerages, mean, they've taken a little bit more of a liberal approach here with letting Manus and these tools access some of these tools, right? [00:14:43] S2: So the brokerages, a [00:14:45] S3: lot