In a recent commentary, Torsten Slok, chief economist at Apollo Global Management, highlighted a new, technology‑driven risk to the banking sector: the emergence of artificial‑intelligence‑powered agents that can autonomously move consumer money from traditional, low‑interest checking accounts to more lucrative alternatives. While the concept may sound like something out of a science‑fiction novel, the underlying mechanics are rooted in real, observable trends in fintech, data analytics, and the increasing sophistication of algorithmic decision‑making tools. At its core, the concern revolves around the speed and scale at which AI agents can act. Traditional bank runs—historical episodes where depositors rush to withdraw cash en masse—were limited by human communication channels, such as word of mouth or newspaper reports.
Today, a single piece of information can be disseminated instantly across millions of devices, and AI agents can interpret that information, assess individual account conditions, and execute transactions in milliseconds. If a credible signal emerges that a particular bank’s yields are sub‑par, an AI‑driven system could automatically reallocate funds to higher‑yielding accounts, money‑market funds, or even emerging digital‑asset platforms, all without the depositor’s direct intervention. The mechanics of such a system are already in place in many financial‑technology applications.
Robo‑advisors, for instance, continuously monitor market conditions and adjust portfolio allocations on behalf of clients. Similarly, some budgeting apps now integrate with users’ bank accounts to suggest optimal cash‑flow strategies, often recommending the transfer of idle balances into savings products that promise better returns. What Slok warns about is the next evolutionary step: agents that not only suggest but also execute these transfers autonomously, driven by pre‑programmed objectives such as maximizing yield or minimizing risk.
Several factors amplify the potential impact of these agents. First, the sheer volume of low‑interest deposits in the United States is enormous.
According to recent Federal Reserve data, household checking accounts collectively hold trillions of dollars, many of which earn negligible interest—often less than 0.1 percent. For many consumers, the opportunity cost of keeping money in such accounts is substantial, especially in a macroeconomic environment where alternative products—high‑yield savings accounts, short‑term certificates of deposit, or even decentralized finance (DeFi) protocols—offer returns that are multiple times higher. Second, the regulatory landscape has historically treated checking accounts as relatively stable funding sources for banks, providing a cheap and reliable base of liquidity. If AI agents begin to systematically drain these deposits, banks could face a rapid contraction in their low‑cost funding, forcing them to rely more heavily on expensive wholesale funding or to raise rates on deposits—a move that could erode profit margins.
Third, the network effect inherent in algorithmic decision‑making can create feedback loops. Imagine a scenario where an AI agent detects a modest uptick in the average yield offered by a competitor’s savings product.
The agent reallocates a portion of its client’s balance to capture the higher return. The collective action of thousands of similar agents amplifies the shift, prompting other institutions to adjust their rates. This dynamic can lead to a cascade of rebalancing events, where the original catalyst—perhaps a small policy change or a market rumor—gets magnified into a systemic liquidity shock. Slok’s warning also touches on the psychological dimension of AI‑driven behavior.
Human depositors often rely on heuristics and social proof; they may be swayed by seeing peers move money or by news of a bank’s declining interest rates. AI agents, however, operate on objective criteria, removing the emotional hesitation that might otherwise slow a run.
This can result in a more abrupt and less predictable outflow of funds, challenging banks’ traditional stress‑testing models that assume a gradual withdrawal pattern. To mitigate these emerging risks, several policy and industry responses are being discussed. Regulators could require banks to disclose the composition of their deposit base in greater detail, including the proportion that is potentially vulnerable to automated reallocation. Enhanced liquidity coverage ratios that specifically account for the speed of AI‑driven withdrawals might be instituted.
On the industry side, banks are already exploring ways to make their own deposit products more attractive to algorithmic agents—by offering tiered interest rates, integrating APIs that allow seamless interaction with fintech platforms, or providing real‑time yield adjustments. Moreover, there is a growing conversation about the ethical design of AI agents themselves. Developers could embed safeguards that limit the frequency or volume of automated transfers, or that require explicit user confirmation for large movements.
Transparency mechanisms—such as audit trails that show when and why an AI agent initiated a transfer—could also help both consumers and regulators monitor systemic exposure. In summary, while the notion of AI agents precipitating a modern bank run may initially sound speculative, the convergence of low‑interest deposit pools, high‑yield alternatives, and increasingly autonomous financial algorithms creates a plausible pathway for rapid, large‑scale fund migrations. Torsten Slok’s cautionary note serves as a call to action for banks, policymakers, and fintech developers alike to anticipate these dynamics, adapt risk frameworks, and ensure that the benefits of AI‑enhanced financial services do not inadvertently destabilize the foundational liquidity that underpins the banking system.