In a recent commentary, Torsten Slok, chief economist at Apollo Global Management, sounded an alarm about the emerging risk that sophisticated artificial‑intelligence agents could inadvertently trigger a new kind of bank run. While traditional runs have historically been driven by panic, rumor, or sudden loss of confidence, Slok points out that the next wave may be orchestrated by algorithms that continuously scan the market for the best return on cash and then act on behalf of their owners without human intervention.
At the heart of the concern is the way modern banking products have evolved. Many consumers keep a substantial portion of their liquid assets in everyday checking accounts that typically earn little or no interest. These accounts are prized for their convenience and safety, but they also represent a low‑yield holding of cash that, in a low‑rate environment, can feel like a missed opportunity.
Over the past few years, a growing number of fintech platforms, money‑market funds, high‑yield savings accounts, and even decentralized finance protocols have begun to offer rates that are several times higher than those provided by traditional banks. Enter AI‑driven financial agents. These are software programs, often built on large‑language models or reinforcement‑learning frameworks, that can be programmed to manage personal finances, optimize investment portfolios, or simply execute routine transactions.
When equipped with access to a user’s bank credentials and a set of pre‑defined objectives—such as “maximize interest earnings while maintaining liquidity”—the agent can monitor interest rates across dozens of institutions in real time. If it detects that a checking account is earning 0.01 % while a competing online savings product offers 4.5 %, the agent can automatically initiate a transfer of funds to capture the higher yield.
On the surface, this behavior seems beneficial for individual savers. However, Slok warns that when millions of households adopt similar agents, the aggregate effect could be dramatic. A modest shift of even 5 % of total checking balances from traditional banks to alternative platforms would represent billions of dollars moving out of the core banking system in a short period.
Because banks rely on these deposits not only for funding loans but also for maintaining required reserve ratios, a rapid outflow could force them to sell assets at a loss, tighten credit, or raise rates to attract new deposits—actions that could ripple through the broader economy. The risk is amplified by the speed and automation of the process. Human‑driven withdrawals are typically staggered, allowing banks time to respond. AI agents, by contrast, can execute thousands of transactions simultaneously, especially if they are linked to a common cloud‑based service or API.
A single software update that adjusts the threshold for “acceptable” interest rates could trigger a cascade of transfers across the entire user base within minutes. This scenario resembles a digital flash crash, but instead of equity prices, the shock would be felt in the liquidity pools of commercial banks. Regulators are beginning to take notice. The Federal Reserve and the Office of the Comptroller of the Currency have issued guidance on the use of third‑party fintech services, emphasizing the importance of robust risk‑management frameworks.
Yet the rapid pace of AI innovation often outstrips the ability of policymakers to craft detailed rules. Slok suggests that a proactive approach is needed, including: 1. **Enhanced Disclosure**: Financial institutions should clearly communicate the interest rates, fees, and liquidity terms of their products so that AI agents can make fully informed decisions without relying on ambiguous marketing language. 2.
**Algorithmic Audits**: Companies that develop or deploy financial AI agents could be required to undergo periodic audits that assess the systemic impact of their algorithms, similar to stress‑testing for banks. 3. **Real‑Time Monitoring**: Central banks could implement dashboards that track large‑scale movements of deposits across institutions, flagging abnormal spikes that may indicate coordinated algorithmic activity. 4.
**Safeguard Mechanisms**: Banks might embed throttling mechanisms that limit the volume or frequency of automated transfers, giving them a buffer to manage liquidity. Beyond the immediate banking sector, the phenomenon raises broader questions about the role of AI in everyday financial decision‑making. As consumers increasingly delegate routine tasks to algorithms—ranging from bill payment to investment rebalancing—the line between personal agency and automated execution blurs. While many users appreciate the convenience and potential cost savings, they may also be unaware of the collective externalities their individual choices generate.
Slok also notes that the problem is not limited to traditional banks. High‑yield alternatives, such as money‑market funds or crypto‑based savings platforms, can also be vulnerable to sudden inflows that strain their operational capacity or expose them to regulatory scrutiny.
A synchronized influx of capital could force these platforms to raise rates unsustainably, leading to a subsequent correction that harms investors. In conclusion, the advent of AI agents capable of autonomously optimizing cash placement represents a double‑edged sword.
On one hand, they empower consumers to extract more value from idle funds, promoting financial efficiency. On the other, they introduce a novel systemic risk that could manifest as a rapid, algorithm‑driven bank run.
Policymakers, financial institutions, and technology developers must collaborate to design safeguards that preserve the benefits of automation while mitigating the potential for destabilizing liquidity shocks. By establishing transparent standards, monitoring aggregate behavior, and building resilience into the banking infrastructure, the financial system can adapt to the era of AI‑enhanced money management without compromising stability.