1. Why are banks blocking AI agents?
Jim Bianco, president of Bianco Research, said during a live stream on October 26 that banks refusing to let outside AI agents into customer accounts are protecting their own revenue instead of their clients. His argument centers on about $4 trillion in deposits that earn customers nothing, income that automated tools could redirect into money market funds. JPMorgan Chase is the most prominent bank enforcing such a ban.
Bianco framed the dispute as a question banks should have to answer in public: whether the restriction exists to shield customers from harm or to preserve profit. Banks that block agents describe the policy as a safety measure. Bianco's reading is that the safety language covers a financial motive, and that the motive becomes clear once the economics of idle cash are laid out.
An AI agent, in this context, is software that a customer authorizes to log into accounts and carry out tasks on their behalf, such as comparing rates, paying bills or moving money between products. The banks' objection is to agents operated by outside companies, which would sit between the customer and the institution that holds the account.
The stakes extend beyond one commentator's view. Banks, brokerages and card issuers are all deciding how much autonomy to grant software that customers increasingly want to use, and each decision carries consequences for security, liability and revenue at the same time. Bianco's contribution to the debate is to insist that the revenue question be named openly rather than left behind the language of risk management.
2. Which banks have restricted outside agents?
JPMorgan Chase prohibits third-party AI agents from signing in to Chase accounts or acting on a customer's behalf. Chief Executive Jamie Dimon explained the policy to analysts on October 13, warning about what he called rogue agents and describing outside agents as too risky for both clients and the bank. A JPMorgan spokesperson said protecting customers is the bank's top priority and that it supports data sharing authorized by customers through open banking providers.
Other large lenders have moved in the same direction. Citi, Bank of America and Wells Fargo have each taken steps to limit third-party agents, and Charles Schwab Chief Executive Rick Wurster has said the brokerage is blocking outside agents that try to move funds within customer accounts.
Wells Fargo has put its approach into contract language. A revised online access agreement takes effect on November 24. Under it, a transaction that a customer authorizes through an AI agent is treated as the customer's own, and the customer bears liability for anything the agent does that was not intended, until the customer revokes the agent's access with the bank. The wording shifts the risk of agent errors onto the account holder rather than prohibiting agents outright.
The differences among these policies are instructive. A flat prohibition, as at JPMorgan, removes the question of agent liability by refusing to recognize agents at all. A contractual approach, as at Wells Fargo, accepts that customers will use agents and assigns the consequences to them. Schwab's stance targets a specific behavior, the movement of funds, rather than every kind of agent activity. Together they show a sector still searching for a standard, with no shared rulebook yet in place.
3. A model built on friction
Bianco's central claim is that traditional finance runs on customer inertia. Most people do not log in every day to shift spare cash from a checking account into a money market fund, even when the fund pays a meaningful yield. An agent would do it without being reminded, and would do it continuously. Banks earn money on deposits that pay their owners nothing, and Bianco said they fear losing that income now that software can automate the transfer.
The scale of the balances involved explains the concern. Bianco put the figure for zero-yield bank balances at about $4 trillion. Federal Deposit Insurance Corporation data in its Quarterly Banking Profile show non-interest-bearing deposits in the domestic offices of U.S. banks at about $4.04 trillion at the end of June 2026, roughly 21% of domestic deposits. That category includes the operating accounts of businesses as well as household savings, so it is not all idle consumer cash.
The same data show how exposed the largest lender is. JPMorgan Chase Bank held $628.8 billion in non-interest-bearing deposits, about 28% of its domestic deposits. Money that pays no interest is an inexpensive source of funding for a bank, which can lend or invest it at higher rates and keep the difference.
A money market fund is a pooled investment that holds short-term, high-quality debt and typically passes most of the interest it earns on to investors. When interest rates are well above zero, the gap between what a fund pays and what a non-interest-bearing account pays can be large. That gap is the reason Bianco sees agents as a threat: they close the distance between a customer's idle cash and the better-yielding option without requiring the customer to act.
Banks measure the profitability of this arrangement through their net interest margin, the difference between what they earn on loans and securities and what they pay for funding. Deposits that cost nothing to hold push that margin up. If a meaningful share of such balances moved elsewhere, banks would have to replace the funding with deposits that pay interest or with wholesale borrowing, both of which cost more. That is the mechanism behind Bianco's claim that the policy defends earnings.
4. What does the credit card comparison show?
Bianco drew a parallel with credit cards. Issuers profit when cardholders pay annual fees but do not fully use the rewards they are entitled to. A customer who leaves points unclaimed or ignores a travel credit is, in effect, subsidizing the program. Some consumers now deploy AI agents to extract the maximum value from those rewards, which narrows the gap that issuers rely on.
The comparison points to a general pattern. Financial products are often priced on the assumption that a share of customers will be inattentive. When software removes the inattention, the pricing logic weakens. In Bianco's view, deposits and card rewards are two examples of the same dynamic, and banks respond to it by controlling who can access accounts.
The analogy also highlights who bears the cost of inertia. In both cases, the customers who pay attention capture value and those who do not leave it with the institution. Agents, by acting on behalf of every user who adopts them, turn attentiveness into something anyone can have, which is why Bianco expects the institutions that benefit from the status quo to resist.
5. Will customers push back?
Bianco expects customers to demand that their agents be allowed in. He noted that people already trade privacy for convenience across much of their online lives, from search to shopping to navigation, and he predicts public pressure on banks will grow as agent tools become more capable and more widely used.
That pressure would likely land on a financial system that has already been working through questions of data access. Open banking arrangements let customers authorize providers to read their account data, which is the route JPMorgan says it supports. The open question is whether agents that can act, and not merely read, should be treated the same way.
Regulators have not settled how agent access should be treated either, and the rules that govern customer data sharing were written before autonomous software became a practical concern. Any attempt to require banks to admit agents would have to address authentication, the recording of customer consent and the allocation of losses when an agent errs or is manipulated. These are the same issues the banks cite as the reason for caution, which means the two sides of the argument overlap more than the rhetoric suggests.
6. A counterpoint on deposit flight
Bianco's co-host Ram Ahluwalia, founder and chief executive of Lumida Wealth Management, agreed that the banks are protecting themselves. He described agents as acting like fiduciaries for the customers they serve, meaning they are supposed to work in the customer's interest.
Ahluwalia also cautioned against overstating how quickly deposits could leave banks. Friction still limits switching, he said, and opening a new bank account is one example. Moving money into a money market fund generally requires a brokerage or similar account, and a customer who has none must first set one up. Those steps slow the flow of deposits even when an agent is willing to do the work.
The exchange between the two hosts therefore leaves a measured conclusion. The incentive Bianco describes is real and can be measured in the deposit figures, but the speed at which it translates into lost balances depends on how easy it becomes for customers to hold accounts at more than one institution and how willing their agents are to use them.
7. What to watch next
The next concrete date is November 24, when Wells Fargo's revised online access agreement takes effect and its liability terms for agent-authorized transactions begin to apply. Whether other banks adopt similar language, or choose outright bans like the one at JPMorgan, will show how the industry intends to handle customers who want automation.
The larger test is whether customer demand overcomes institutional reluctance. If agents become a routine way to manage cash, the roughly $4 trillion in non-interest-bearing balances that Bianco highlighted is the pool most likely to feel it, and the banks' answer to the question he posed will matter well beyond this single dispute.
For customers, the practical implication is to read the terms of any account agreement before authorizing an agent, since the wording on liability may change who pays when something goes wrong. For banks, the choice is between competing on yield for deposits that currently cost them nothing and continuing to restrict the tools that make such competition easier. Which course prevails will become clearer as more institutions publish their own agent policies in the coming months.