In a recent commentary, Torsten Slok, a senior economist at Apollo Global Management, sounded the alarm on a new, technology‑driven risk to the stability of the banking sector. While traditional bank runs have historically been sparked by panic, rumor, or sudden loss of confidence, Slok suggests that the next wave could be orchestrated by sophisticated artificial‑intelligence agents that act on behalf of consumers.
These AI‑powered tools, designed to optimize personal finances, could systematically and automatically move money out of low‑interest checking accounts and deposit it into higher‑yield alternatives such as money‑market funds, short‑term bonds, or even crypto‑based savings platforms. The cumulative effect of millions of such agents acting in concert could resemble a classic bank run, but without the emotional contagion that typically fuels one. ### Why AI Agents Might Target Low‑Yield Deposits The modern consumer faces a bewildering array of financial products, each promising a different blend of safety, liquidity, and return.
Traditional checking accounts, especially those offered by large, legacy banks, often provide negligible interest—sometimes zero percent—while charging fees for services that many customers no longer need. Meanwhile, fintech firms and online banks have introduced high‑yield savings accounts that can offer rates several times higher, and investment platforms provide easy access to short‑term, low‑risk instruments that generate modest but meaningful returns. AI agents are increasingly embedded in personal finance apps, robo‑advisors, and even voice‑activated assistants. Their core function is to analyze a user’s cash flow, spending patterns, and financial goals, then recommend—or automatically execute—transactions that improve the user’s net worth.
When these agents detect that a checking account is earning less than the market rate for comparable liquidity, they may initiate a transfer to a more attractive vehicle. This behavior is rational from an individual standpoint: every dollar earns a higher return, and the user’s overall financial health improves.
### The Scale of Potential Outflows If a single AI agent moves a few hundred dollars, the impact on any one bank is negligible. However, Slok points out that the proliferation of these agents across millions of households could generate a massive, coordinated shift of capital. Consider a scenario in which 10 million households each have an average of $5,000 in low‑interest checking balances. If AI agents automatically reallocate just 20 % of those funds to higher‑yield accounts, the banking system would see an outflow of $10 billion in a relatively short period.
While large banks have substantial reserves, a rapid, simultaneous withdrawal of this magnitude could strain liquidity buffers, force banks to sell assets at unfavorable prices, or trigger tighter credit conditions. ### Differences From Traditional Bank Runs Traditional runs are often precipitated by rumors or a loss of confidence, leading depositors to physically line up at branches. The modern version envisioned by Slok would be digital, silent, and algorithmic.
There would be no crowds at teller windows; instead, a surge of electronic transfers would appear on a bank’s balance sheet. This could catch risk managers off guard because the outflows would be systematic rather than random, reflecting a common decision‑making logic embedded in the AI code. Moreover, the speed of execution is dramatically faster. An AI agent can initiate a transfer the moment it detects a rate disparity, potentially completing the move within seconds.
In contrast, human‑driven runs unfold over hours or days, giving banks time to respond. The algorithmic nature also means that once a trigger is set—such as a specific interest‑rate threshold—every participating agent will act simultaneously, amplifying the shock. ### Potential Mitigations and Policy Responses Slok recommends that regulators and banks proactively address this emerging risk. First, banks could enhance the attractiveness of their own deposit products by offering tiered interest rates that reward larger balances or longer holding periods, thereby reducing the incentive for automatic reallocation.
Second, transparency is key: financial institutions should clearly disclose how their rates compare to market alternatives, helping AI agents make more balanced decisions. Regulators might consider imposing limits on the speed or volume of automated transfers, similar to the safeguards that exist for high‑frequency trading in securities markets. For example, a cooling‑off period could be introduced, requiring a short delay before large, automated transfers are executed. Additionally, banks could be required to maintain higher liquidity ratios specifically earmarked for sudden, algorithm‑driven outflows.
From a technology‑governance perspective, developers of personal‑finance AI tools should embed safeguards that prevent mass, simultaneous actions based on a single market signal. This could involve randomizing the timing of transfers, setting caps on daily movement amounts, or incorporating a risk‑adjusted decision framework that weighs the systemic impact of a transfer against the individual benefit. ### The Broader Implications for the Financial System If left unchecked, AI‑induced deposit migrations could accelerate the shift away from traditional banking toward a more fragmented, fintech‑driven landscape. While competition can drive innovation and better rates for consumers, it also raises questions about the resilience of the banking system.
Deposit insurance schemes, such as the FDIC in the United States, are designed to protect against panic‑driven runs, but they may need to be reevaluated in the context of algorithmic behavior that can bypass the emotional triggers the insurance is meant to mitigate. Furthermore, the concentration of deposits in a few high‑yield platforms could create new systemic risks.
If a large, tech‑focused savings platform were to experience a cyber‑attack or operational failure, the fallout could ripple across the broader economy, similar to the contagion effects observed during the 2008 financial crisis. ### Conclusion Torsten Slok’s warning underscores a subtle yet potent vulnerability: the same technology that helps individuals optimize their finances could, at scale, destabilize the institutions that underpin the economy. By recognizing the potential for AI agents to orchestrate a digital, automated bank run, banks, regulators, and fintech developers have an opportunity to design safeguards before the risk materializes. Proactive adjustments to deposit pricing, enhanced transparency, regulatory limits on automated transfers, and responsible AI design can collectively mitigate the threat.
As the financial ecosystem continues to evolve, balancing innovation with stability will be essential to ensure that the benefits of AI‑driven personal finance do not inadvertently undermine the very foundations of the banking system.