In a recent commentary, Torsten Slok, chief economist at Apollo Global Management, sounded the alarm on a potential new threat to the stability of the banking sector: the rise of autonomous artificial‑intelligence agents that can move money at a speed and scale previously unimaginable. While most discussions about AI focus on its impact on jobs, productivity, or data privacy, Slok is drawing attention to a more subtle but equally consequential risk—an AI‑induced bank run.
Traditional bank runs have historically been triggered by panic: depositors rush to withdraw cash when they fear a bank is insolvent, creating a self‑fulfilling prophecy. In the digital age, the mechanics of a run have changed. Most consumers keep their money in electronic accounts rather than holding physical currency, and withdrawals are processed instantly through online banking apps. This means that a wave of panic can now be transmitted across the entire banking network in a matter of minutes, if not seconds.
What makes the current situation different, according to Slok, is the emergence of sophisticated AI agents that can act on behalf of individuals or institutions without direct human oversight. These agents are already being used to optimize personal finances—automatically paying bills, reallocating investments, and finding the best interest rates for savings. When programmed to prioritize the highest possible return on idle cash, these agents naturally scan the market for alternatives that outperform the modest yields offered by traditional checking accounts.
In a low‑interest environment, many banks provide checking accounts that pay virtually nothing on deposits, relying instead on the volume of funds to generate income through loans and other services. Meanwhile, a growing number of fintech platforms, money‑market funds, and digital wallets are offering substantially higher yields, often backed by short‑term government securities or other low‑risk assets.
An AI agent tasked with maximizing a user’s net interest earnings will flag these opportunities and, if instructed, move the funds automatically. The danger arises when a critical mass of such agents make the same decision at roughly the same time.
Because the agents operate on similar data sets—interest rates, yield curves, and risk assessments—they are likely to converge on the same set of high‑yield alternatives. If millions of households each have an AI‑driven “cash optimizer” that decides to pull money out of a low‑interest checking account and deposit it into a higher‑yield vehicle, the aggregate outflow could be massive and abrupt.
Unlike a traditional run, where human emotions and rumors drive behavior, an AI‑driven exodus would be deterministic and algorithmic. It would be triggered by measurable changes in rates rather than speculation about a bank’s solvency. This could lead to a scenario where a bank experiences a sudden, large‑scale reduction in deposits without any underlying financial distress, simply because its products are no longer the most attractive from a purely financial‑return perspective.
The implications for banks are profound. Deposits are a primary source of cheap funding that banks use to lend to businesses and consumers. A rapid decline in deposit balances forces banks to seek more expensive funding sources, such as wholesale borrowing or issuing debt, which can erode profit margins and increase risk exposure.
Moreover, the perception that a bank is losing deposits—even if the loss is algorithmic—can undermine confidence and potentially trigger a feedback loop where other depositors, whether human or AI‑driven, also withdraw funds. Regulators have begun to consider how to address this emerging risk. Some proposals include requiring banks to disclose more detailed information about the interest rates and terms of their deposit products, thereby reducing information asymmetry that AI agents exploit.
Others suggest implementing “cool‑down” periods for large automated transfers, similar to the settlement delays used in securities markets to prevent flash crashes. Additionally, central banks could develop guidelines for the responsible design of financial‑optimizing AI, encouraging transparency in the decision‑making criteria used by these agents. From a technological standpoint, there are also opportunities to mitigate the risk.
Banks could integrate their own AI tools that monitor real‑time deposit flows and predict potential outflows before they become critical. By offering competitive, dynamically adjusted rates or bundled services that add value beyond pure yield, banks can make their accounts more resilient to purely algorithmic arbitrage.
Consumers, too, have a role to play. While AI agents can enhance financial outcomes, users should remain aware of the broader systemic impact of their automated decisions.
Setting parameters that limit the frequency or volume of transfers, or opting for a diversified mix of accounts, can help prevent unintended concentration of withdrawals from a single institution. In summary, Torsten Slok’s warning highlights a novel vector for financial instability that stems not from panic, but from the very efficiency and rationality that AI promises.
As AI agents become more capable of managing personal finances, the collective actions they take can inadvertently create a modern form of a bank run—one driven by algorithmic optimization rather than fear. Stakeholders across the financial ecosystem, including banks, regulators, technology developers, and consumers, must collaborate to design safeguards that preserve the benefits of AI while protecting the stability of the banking system.