No loss of confidence, no queues outside bank branches, no panic withdrawals: The next threat to banks’ deposits could look remarkably uneventful. Software checks a user’s accounts, finds a better interest rate elsewhere and automatically moves excess cash. For customers, that would be efficient. For banks, it could become expensive.
The money that simply stays put
Banks do not make money solely from whom they lend to and the interest rates they charge. Just as important is how cheaply they can fund themselves.
Customer deposits play a central role. Checking and traditional savings accounts in particular provide many banks with relatively inexpensive funding. While loans and other assets generate higher returns, customers often receive little interest on at least part of their balances. The difference contributes to banks’ net interest margins.
Artificial intelligence could attack precisely that mechanism.
At the end of September, Torsten Slok, chief economist at asset manager Apollo Global Management, raised the prospect of an “agentic bank run.” His argument: Personal AI agents could eventually monitor household accounts and automatically move surplus cash wherever higher yields are available.
The difference can be substantial. According to data cited by Slok, the average interest rate on US checking accounts is only around 0.1 percent, while various online banks and financial platforms offer between 3.3 and five percent.
As of early October, some high-yield savings accounts in the United States offer rates of around 4.5 percent. The nationwide average for traditional savings accounts, by contrast, remains below 0.4 percent.
Consumers have been able to exploit that gap for years. Yet many rarely switch accounts. Effort, habit and the convenience of keeping several financial services with one provider have so far protected banks from constant competition for every basis point.
An AI agent could remove exactly that friction.
From recommendation to autonomous action
Digital financial services have so far mainly helped customers compare options. Users still have to search for offers, make decisions and initiate transfers themselves.
Agentic AI is designed to go a step further. Such systems do not merely answer questions; they can carry out multi-step tasks autonomously and interact with other applications on behalf of their users.
Meta launched Muse, its personal AI agent, in early September. In financial services, Muse can connect bank accounts via Plaid, analyse statements, create budgets, identify recurring expenses and, with the user’s approval, cancel unused subscriptions.
Automatically moving deposits to whichever bank currently offers the highest rate is not among the standard functions Meta currently describes.
Slok’s scenario therefore represents a possible next stage of development – not something millions of Muse users are already doing today.
But the technological direction is clear: The more financial data an agent can analyse and the more transactions it is permitted to execute, the lower the effort required to optimise cash continuously.
Why that would matter for banks
For a single customer account, the effect may appear small. Across the banking system’s entire deposit base, the calculation changes.
If customers demand higher returns on their cash or switch providers more quickly, banks’ funding costs rise. Institutions can respond by offering higher deposit rates themselves. That may retain the deposits, but it compresses net interest margins.
Alternatively, banks may try to compensate for lower margins through higher lending rates, additional fees or cost reductions. Lending standards could also become more selective.
In 2025, banks insured by the Federal Deposit Insurance Corporation generated a combined net income of 295.6 billion dollars. The FDIC also noted that lower funding costs were among the factors supporting profitability.
Persistent competition for deposits would put pressure on exactly that side of the balance sheet.
Slok therefore warns that banks could lose a substantial share of their cheap deposits if households begin using AI agents to optimise returns on cash automatically.
That scenario, however, differs fundamentally from a traditional bank run.
Not a bank run in the 2008 sense
In a conventional bank run, customers withdraw money because they fear for an institution’s solvency. The dynamics are driven by panic: Everyone wants access to their money before it may be too late.
The scenario described by Slok involves no such loss of confidence. Cash moves not because customers fear a bank, but because an algorithm identifies a higher return elsewhere.
There is another important distinction. Not every transfer would remove money from the banking system. If a customer merely moves funds from a low-yield bank to one offering higher rates, the deposit shifts within the system. If the money instead flows into money-market funds or other investment products, the impact on bank funding becomes more significant.
The term “bank run” therefore describes mainly the potential speed and scale of the movement – not necessarily a banking crisis in the traditional sense.
The more immediate threat to banks would be persistent pressure on margins.
Markets have already noticed the risk
Investors are no longer treating AI disruption in finance as a purely theoretical experiment.
On 22 September, the S&P 500 financial sector fell two percent, while the S&P banking index dropped three percent. Shares in brokerage firm Charles Schwab declined 6.1 percent, Ameriprise Financial lost 4.4 percent and Raymond James fell more than three percent.
Alongside movements in the bond market, analysts cited growing concerns about AI competition in financial services as one of the pressures on the sector. Meta had launched Muse only two weeks earlier, and the product quickly reached millions of downloads.
The most immediate threat may initially fall on wealth managers, brokers and financial advisers. If AI can analyse portfolios, compare products and automate financial planning, traditional fee models could come under pressure.
For banks, however, the potential disruption goes deeper. The question is not only who gives customers financial advice in future. It is who controls where their money sits.
Not every bank would be affected equally
Large universal banks have one important advantage: They can offer checking accounts, credit cards, brokerage accounts, mortgages and other services within a single ecosystem.
Customers who have several financial products closely linked with one institution may still be reluctant to move money constantly, even if slightly higher rates are available elsewhere.
Other banks depend more heavily on cheap deposits or offer fewer additional services that create customer loyalty.
AI could therefore accelerate a trend that higher interest rates have already set in motion. As banks and money-market funds began offering attractive returns again, customers became more attentive to what happens to idle cash.
The real innovation would not be the existence of better offers.
It would be the automation of switching.
Apollo has an interest in the outcome
Slok’s warning should also not be treated as a neutral academic forecast.
Apollo is one of the world’s largest alternative credit investors. At the end of the second quarter, the company said it managed 849 billion dollars across its credit strategies.
Private credit funds and other capital-market providers have increasingly taken over parts of the financing role once dominated by traditional bank balance sheets.
If banks come under pressure from higher funding costs or reduce lending, alternative lenders can benefit.
That does not invalidate Slok’s argument. But it matters when assessing the perspective from which it is being made.
How quickly AI agents will gain direct access to bank accounts, how many consumers will permit them to execute autonomous transfers and whether regulators will restrict such capabilities remain open questions.
AI could eliminate the inertia premium
For investors, that may be the central issue.
Banks do not own only brands, branches, loan books and technology. They also benefit from customer behaviour. People rarely switch their bank accounts, do not compare interest rates every day and do not optimise every spare dollar.
That inertia has economic value.
AI agents could reduce it because they perform tasks that people could theoretically carry out already – but often do not.
This creates an unusual form of disruption for the financial sector. AI does not necessarily have to build a better bank.
It may be enough if it makes the customer behave more consistently like a rational investor.
For banks, that could become one of the most expensive efficiency gains artificial intelligence ever creates.
Stefanie S. Klief
