In a recent commentary, Torsten Slok, chief economist at Apollo Global Management, sounded the alarm on a potential new catalyst for bank runs that stems not from panic or misinformation, but from the rapid, automated actions of artificial‑intelligence‑driven financial agents. These sophisticated software programs, often referred to as AI agents or autonomous financial bots, are increasingly being integrated into personal finance platforms, wealth‑management apps, and even everyday banking interfaces.

Their primary function is to optimize the allocation of a user’s cash, seeking the highest possible return while minimizing risk. While this capability promises greater efficiency and higher yields for consumers, Slok cautions that it could also create a systemic vulnerability: a swift, coordinated outflow of funds from traditional low‑interest checking accounts into alternative, higher‑yield vehicles such as money‑market funds, short‑term bond ETFs, or decentralized finance (DeFi) protocols.

### How AI Agents Operate Modern AI agents are built on machine‑learning models that continuously ingest market data, interest‑rate changes, and product‑specific terms. When a new, more attractive cash‑management product appears—say, a savings account offering a 4.5% annual percentage yield (APY) compared to the 0.5% APY on a typical checking account—the agent evaluates the marginal benefit of moving a portion or the entirety of the balance.

It then executes the transfer automatically, often without requiring the account holder to approve each individual transaction. In many cases, the user has pre‑authorized the agent to act on their behalf under a set of predefined rules, such as “maintain a minimum balance of $1,000 in the checking account, otherwise invest excess cash in the highest‑yielding short‑term instrument.” These agents are not limited to a single bank or platform.

Through open‑banking APIs, they can access accounts across multiple institutions, compare rates in real time, and shift money wherever the best return is available. The process is frictionless: a few clicks to grant permission, and the AI takes over the day‑to‑day cash‑management decisions.

### The Potential for a Digital Bank Run Historically, bank runs have been driven by depositor fear—often sparked by rumors, news of insolvency, or a loss of confidence in a bank’s ability to meet withdrawal demands. In the digital age, the fear factor could be replaced by algorithmic efficiency.

If a large number of AI agents receive the same signal—such as a central bank’s decision to lower the federal funds rate, thereby reducing the yield on traditional checking accounts—their programmed logic may simultaneously trigger mass transfers out of those accounts. Unlike a human‑driven run, which unfolds over hours or days as news spreads, an AI‑driven exodus could happen within minutes or even seconds, as the agents execute trades in parallel.

The scale of the risk is amplified by the sheer volume of low‑interest deposits that exist in the United States. According to Federal Reserve data, households hold trillions of dollars in checking and transaction‑type accounts that earn negligible interest. If even a modest percentage of these balances were reallocated by AI agents to higher‑yield alternatives, the resulting liquidity drain could strain banks that rely on stable deposit bases to fund loans and other operations. ### Why Banks Might Be Vulnerable Traditional banks maintain a fraction of deposits as reserves, typically a small percentage dictated by regulatory requirements.

The remainder is lent out or invested in longer‑term assets. A sudden, large‑scale withdrawal forces banks to liquidate assets, often at unfavorable prices, to meet the demand for cash.

In a scenario where AI agents simultaneously pull funds from multiple banks, the pressure on the interbank market could intensify, raising short‑term borrowing costs and potentially leading to a cascade of liquidity shortages. Furthermore, many banks have not yet modernized their technology stacks to handle real‑time, high‑frequency deposit movements. Legacy systems may be unable to process a surge of automated transfers efficiently, leading to processing delays, errors, and further erosion of depositor confidence. ### Mitigation Strategies Slok suggests that regulators and financial institutions should begin to consider the systemic implications of AI‑driven cash management.

Potential measures include: 1. **Enhanced Disclosure**: Requiring fintech platforms and banks to disclose the extent to which AI agents can move funds on behalf of customers, and under what conditions.

2. **Rate‑Lock Mechanisms**: Allowing customers to lock in a rate for a specified period, preventing automatic migration until the lock expires. 3. **Liquidity Buffers**: Encouraging banks to hold larger liquidity cushions to absorb sudden outflows without resorting to fire‑sale asset disposals.

4. **Stress Testing**: Incorporating AI‑driven deposit migration scenarios into regular stress‑test frameworks, similar to how banks test for interest‑rate shocks and credit losses. 5. **Regulatory Oversight of APIs**: Monitoring open‑banking APIs to ensure they include safeguards that limit the speed or volume of automated transfers.

### The Broader Context The issue fits within a larger narrative about the unintended consequences of financial technology. While AI agents can democratize access to higher yields—especially for individuals who lack the time or expertise to shop around for better rates—they also introduce a new form of coordination risk.

In the same way that high‑frequency trading has reshaped market dynamics in equities, AI‑driven cash management could reshape the dynamics of bank deposits. It is also worth noting that the problem is not limited to traditional banks.

Money‑market funds, online‑only banks, and even crypto‑based stablecoin platforms could become the destination for the redirected cash. This could create competitive pressures that further erode the market share of legacy institutions, accelerating the shift toward a more fragmented, technology‑centric banking ecosystem. ### Conclusion Torsten Slok’s warning serves as a reminder that innovation, while beneficial, can also generate systemic risks when it changes the speed and scale at which financial decisions are made.

As AI agents become more prevalent in everyday banking, regulators, banks, and fintech firms must collaborate to design safeguards that preserve liquidity stability while still delivering the promised efficiency gains to consumers. By anticipating the possibility of an AI‑driven bank run and implementing proactive measures, the financial system can better navigate the transition toward an increasingly automated future.