In a recent commentary, Torsten Slok, the chief economist at Apollo Global Management, highlighted a growing concern among financial regulators and market participants: the potential for artificial‑intelligence‑driven agents to destabilize the banking system by rapidly moving deposits out of traditional, low‑interest accounts. While the concept may sound futuristic, the underlying mechanics are rooted in existing financial technology trends and the ever‑increasing sophistication of algorithmic decision‑making tools. At its core, the warning revolves around the way modern AI agents are programmed to optimise financial outcomes for their users. These agents—whether embedded in personal finance apps, robo‑advisors, or integrated directly into banking platforms—continuously scan a multitude of data points, from interest rate changes and inflation expectations to macro‑economic indicators and individual spending habits.
When they detect a more attractive yield elsewhere—be it a high‑interest savings account, a short‑term Treasury bill, a money‑market fund, or even a decentralized finance (DeFi) protocol—they can automatically execute the transfer of funds without requiring explicit, manual approval from the account holder. In a low‑interest environment, many households keep large sums of cash in checking accounts that earn little to no return.
Traditionally, moving that cash required a conscious decision: the account holder would log into their online banking portal, fill out a transfer form, and confirm the transaction. The friction involved in this process acted as a natural brake on rapid, large‑scale outflows. However, AI agents eliminate much of that friction. By leveraging APIs provided by banks and fintech platforms, they can trigger transfers in seconds, often at the most opportune moment—such as the instant a new, higher‑yield product is announced or when market conditions suggest a temporary spike in returns.
Slok’s concern is that, if a sizable portion of the population adopts such agents, the aggregate effect could resemble a classic bank run, albeit one driven by algorithmic logic rather than panic. A bank run typically occurs when depositors, fearing insolvency, rush to withdraw their money, forcing the institution to liquidate assets at fire‑sale prices and potentially leading to a solvency crisis.
In the AI‑driven scenario, the trigger is not fear but optimisation. The agents collectively respond to a signal—say, a 0.5‑percentage‑point increase in the rate offered by a competitor—and simultaneously shift billions of dollars out of checking accounts.
The speed and coordination afforded by machine‑to‑machine communication could overwhelm a bank’s liquidity management systems before it has a chance to rebalance its asset portfolio. Several factors amplify the risk.
First, the sheer scale of data integration means that once a new, higher‑yield product is listed, it can be disseminated across millions of devices in real time. Second, many AI agents operate under similar optimisation criteria—maximising net interest margin for the user—so their behavior converges, creating a herd effect.
Third, regulatory frameworks have historically been designed around human‑initiated transactions, with safeguards such as daily transfer limits and manual verification steps. These safeguards may be insufficient when the trigger is an autonomous script that can bypass or exploit loopholes in the system.
The potential consequences extend beyond the immediate liquidity strain on banks. A rapid outflow of deposits can force banks to sell longer‑term securities at a loss, compressing their net interest margins and reducing the capital available for lending. This, in turn, could tighten credit conditions for businesses and consumers, slowing economic growth. Moreover, the perception of vulnerability could lead to a feedback loop: as news spreads that AI agents are moving money en masse, more users might manually withdraw funds, further exacerbating the strain.
To mitigate these risks, Slok suggests a multi‑pronged approach. Regulators could require banks to implement real‑time monitoring of deposit flows, flagging abnormal patterns that may indicate algorithmic activity. Enhanced stress‑testing scenarios that incorporate rapid, large‑scale withdrawals triggered by AI agents would provide a clearer picture of systemic resilience.
On the technology side, banks might impose stricter API access controls, limiting the frequency and volume of automated transfers, or requiring additional authentication for high‑value moves. Financial institutions also have an opportunity to turn this challenge into a competitive advantage.
By offering their own AI‑driven savings products that dynamically adjust rates in response to market conditions, banks can retain deposits while still delivering higher yields to customers. Transparent communication about how AI agents operate and the safeguards in place can build trust and reduce the likelihood of sudden, unanticipated outflows.
In summary, while the notion of AI agents precipitating a bank run may sound like a plotline from a techno‑thriller, the underlying dynamics are firmly grounded in current financial technology trends. Torsten Slok’s warning serves as a timely reminder that the same tools designed to optimise personal finance can, if left unchecked, generate systemic pressures on the banking sector.
Proactive regulatory oversight, robust technological safeguards, and innovative product design will be essential to ensure that the benefits of AI‑enhanced financial management are realised without compromising the stability of the broader financial system.