In a recent commentary, Torsten Slok, chief economist at Apollo Global Management, sounded the alarm on a new, technology‑driven risk to the stability of the banking system. He argued that the rapid advancement of artificial‑intelligence agents—software programs capable of making autonomous financial decisions—could inadvertently trigger a modern‑style bank run.
Unlike the classic panic‑driven runs of the past, this scenario would be engineered by algorithms that constantly scan the market for better yields and then move money from traditional, low‑interest checking accounts into higher‑yield products such as money‑market funds, short‑term bond ETFs, or even crypto‑based savings platforms. Slok’s concerns stem from the observation that today’s consumers are already accustomed to digital tools that optimize their finances. Budgeting apps, robo‑advisors, and personal finance dashboards routinely recommend ways to reduce fees or increase returns.
The next logical step, he suggests, is the deployment of AI agents that not only advise but also execute transactions without human intervention. These agents could be programmed to monitor interest rates in real time, compare the effective annual yield of various deposit vehicles, and automatically shift balances the moment a more attractive option appears.
While such automation promises convenience and higher earnings for individual savers, it also introduces a systemic vulnerability: the collective action of millions of agents acting in unison could drain liquidity from banks at a speed far exceeding that of a traditional panic‑driven withdrawal. To understand the mechanics, consider a typical household checking account that offers a nominal interest rate of 0.01%—a figure that has been effectively zero for years in many jurisdictions. An AI agent programmed to maximize returns would quickly identify that a high‑yield online savings account offers 4% APY, or that a short‑term Treasury fund yields 3.5%.
Within seconds, the agent could initiate a transfer of the household’s idle cash to the higher‑yield product. If a large proportion of the population employs similar agents, the cumulative effect would be a massive outflow of deposits from traditional banks. Because banks rely on these deposits to fund loans and other assets, a sudden, large‑scale reduction in their deposit base could force them to sell securities at a loss, tighten credit, or, in extreme cases, seek emergency liquidity from central banks. Slok points out that the speed and coordination of AI‑driven transfers differ fundamentally from human‑initiated runs.
Historically, a bank run required a rumor or a loss of confidence that spread through word of mouth, newspapers, or later, television. The process was relatively slow, giving banks some time to respond with reassurance or temporary measures. In contrast, AI agents operate on millisecond timescales and can execute thousands of transactions simultaneously across the globe. Moreover, the decision logic embedded in these agents is deterministic: if the yield differential exceeds a predefined threshold, the transfer is triggered.
This eliminates the emotional hesitation that might otherwise slow a human’s decision to withdraw funds. The potential ramifications extend beyond individual banks. A coordinated shift of deposits could reshape the entire financial ecosystem. Traditional banks might see a permanent reduction in low‑cost funding, compelling them to raise interest rates on checking accounts—a move that could erode their competitive advantage and increase the cost of banking services for all customers.
Meanwhile, alternative financial platforms that attract these AI‑directed flows could experience rapid growth, but they may also be less regulated, raising concerns about consumer protection, systemic oversight, and the stability of the broader credit market. Regulators are beginning to take note.
Slok suggests that policymakers should consider new safeguards, such as imposing limits on the frequency or volume of automated transfers, requiring banks to maintain higher liquidity buffers, or developing real‑time monitoring tools that can detect unusual patterns of algorithmic activity. He also recommends that financial institutions enhance the transparency of their product offerings, making it easier for both humans and AI agents to compare yields without resorting to opaque third‑party platforms.
From a technological perspective, developers of AI agents have a responsibility to embed risk‑aware features into their algorithms. For instance, agents could be programmed to stagger transfers over time, incorporate a “cool‑down” period after a large move, or factor in macro‑economic indicators that signal broader market stress. By building in such safeguards, the industry can mitigate the risk of a synchronized exodus of funds while still delivering the promised benefits of higher returns. In summary, Torsten Slok’s warning highlights a paradox of modern finance: the very tools designed to improve efficiency and maximize returns could, if left unchecked, undermine the stability of the institutions that underpin the economy.
As AI agents become more sophisticated and more widely adopted, both banks and regulators must anticipate and address the systemic risks they pose. The challenge will be to strike a balance that preserves the innovative advantages of automation while safeguarding the liquidity and confidence that banks need to function effectively. Only through proactive policy, thoughtful design of AI agents, and ongoing dialogue among stakeholders can the financial system adapt to this emerging frontier without succumbing to a new kind of algorithm‑driven bank run.