Apollo Global Management chief economist Torsten Slok has flagged a new threat to the banking system: the "agentic bank run." In a note published on September 27, Slok argued that if millions of consumers allow AI agents to move their savings in pursuit of higher interest rates, nothing would stop those agents from acting in unison within the same week.

A bank run, in the textbook sense, is depositors withdrawing funds en masse. The version Slok describes swaps panic for yield: instead of fleeing banks they fear are failing, account holders would let software chase better returns — while keeping the defining feature of a run, many deposits moving at once.

How AI agents could banks

Slok wrote that consumer AI agents could soon pull household cash out of low-yield checking accounts and place it into fintech accounts paying far more. A $10,000 checking balance at the 0.1% national average earns roughly $10 a year, while the same deposit in a 5% account would earn $500.

Drawing on data from the FDIC, Haver Analytics and rate information from providers including Revolut, Varo Bank and Wealthfront, Slok compiled a list of 11 fintech and online accounts paying between 3.3% and 5%, led by Adelfi at 5%, with SoFi close behind at 4.5%.

Banks keep checking rates near zero because those deposits are their cheapest source of loan funding, and retail banking models assume most of that money stays put. Slok warns that if AI agents begin moving substantial amounts, banks will lose the cheap deposit base that makes lending profitable. The model depends on inertia — on most depositors never bothering to move their money — and an agent tasked with maximizing yield undercuts that assumption.

The Bank of England voiced a similar concern in its April 2025 Financial Stability in Focus paper, warning that advanced AI trading strategies could push firms into "increasingly correlated positions and acting in a similar way during a stress, thereby amplifying shocks." The paper defines agentic AI as systems that take autonomous action toward set goals by using tools and adapting to their environment.

Which AI agents can make bank transfers?

The trigger cited in Slok's note is Meta's Muse, launched on September 8 with account connections powered by Plaid. Muse can currently view a user's balances, transactions and investment holdings, but it cannot yet initiate transfers.

That read-only limitation is where the debate currently rests: Slok's scenario requires agents that can act on account data, not merely see it.

Muse shipped in the same week that Jacob Coxon, a researcher who had worked at both OpenAI and Anthropic, quit the field and accused the two companies of "gambling with our lives." Cryptopolitan previously reported that Meta's leadership went on to push back against calls for an industrywide slowdown.

By contrast, at the UN Security Council on September 23, Anthropic's Dario Amodei pledged to "slow down as much as necessary," while OpenAI's Sam Altman called for "extreme care."

Warnings about synchronized AI behavior in markets are not new. In a 2023 speech, then-SEC Chair Gary Gensler cautioned that a few dominant AI base models could make "similar decisions because they are getting the same signal." His herding concern is the market-side twin of Slok's deposit-side worry: separately controlled accounts converging on the same move at the same time.

The debate comes against a shifting rate backdrop: the Federal Reserve raised its target range to 3.75% to 4% on September 16, its first hike in three years, and bank stocks fell afterwards as investors worried that a flatter gap between short- and long-term rates would squeeze the profit banks earn on loans. The spread at the center of Slok's argument — 0.1% checking against 3.3% to 5% alternatives — is exactly the kind of discrepancy agents built to pursue higher interest rates would act on.