Apollo’s chief economist warned that AI assistant agentic payments that can move household savings into higher-yielding accounts could trigger what he called an “agentic bank run,” draining deposits banks’ use to fund loans as AI assistants gained more financial control since social media and mobile banking accelerated the 2023 collapse of Silicon Valley Bank.
On September 27, the chief economist published a note, “Is an Agentic Bank Run Coming?” through Apollo’s Daily Spark research series. Apollo Global Management’s Torsten Slok manages around $1 trillion in assets.
It does not describe an AI-driven bank run that has already happened. An International Monetary Fund (IMF) analysis of earlier bank failures helps explain why the speed of withdrawals matters.
What Torsten Slok’s Agentic AI Warning Says
Slok warned that AI agents, such as Meta’s AI Assistant Muse, and similar tools could automatically sweep household cash out of low-interest checking accounts, which currently pay a national average of just 0.1%, and into fintech accounts offering between 3.3% and 5.0% annual interest.
Apollo’s chief economist’s accompanying chart identified 11 fintech and online deposit products paying above the national average, led by Adelfi at 5.0% and SoFi at 4.5%.
For one saver, taking the better rate may be sensible. For a bank serving millions of savers, many simultaneous transfers could create a funding problem.
“If every household used AI agents to optimize the return on their cash balances, banks could lose a large share of the cheap deposits they rely on to make loans,” wrote Slok, highlighting that outcomes from ecommerce agent adoption would trouble the wider financial system.
A Systematic Risk, Not Isolated One
Banks rely on low-cost checking deposits to fund lending at scale – an arrangement dependable on most depositors leaving that cash largely untouched. They depend on deposits remaining relatively stable. If customers transfer large amounts quickly, a bank may need to find replacement funding or sell assets before it can comfortably meet demand.
A higher savings rate elsewhere does not mean the original bank is failing, but widespread transfers can still put pressure on it.
The same convenience behind agentic shopping explains the concern: software can compare options and act faster than someone checking each offer manually.
An ecommerce agent might find a cheaper product; a finance assistant could find a better rate. The comparison is about speed and delegation, though a bank transfer carries different consequences from a purchase.
The distinction matters for AI shopping agents, which may be allowed to buy items within a user’s spending rules.
The potential risk from agentic payments arises if similar authority extends to moving savings. Whether an assistant can transfer money without asking each time depends on the permissions a customer gives it and the controls imposed by the account provider. Slok’s note describes a possible capability, not proof that Meta’s Muse currently moves deposits on its own.
How Banks Could Respond
The IMF’s authors examined the 2023 failures of Silicon Valley Bank, Signature Bank and First Republic Bank, as well as Credit Suisse.
They said mobile banking and the spread of information through social media helped customers withdraw money rapidly. Those were technology accelerated runs, but the analysis does not identify them as runs caused by autonomous AI assistants.
“Fast withdrawals, on the contrary, can bring down a bank before it can secure funding alternatives,” the authors wrote.
They warned that future AI tools might compare rates, assess perceived safety and reallocate deposits according to criteria set by customers. That makes agentic payments a question for financial stability as well as convenience.
For banks, agentic AI payments would make withdrawal patterns harder to predict if many assistants followed similar instructions at once. The IMF suggested using technology to monitor those patterns and improve liquidity management. Banks could also hold more readily available assets or strengthen other sources of funding, giving them more room to meet sudden transfers.
Regulators have further options, including deposit insurance, access to central bank funding during a crisis, and arrangements for handling a bank failure. Each has costs or possible side effects. The IMF cautioned against assuming that one safeguard could remove every risk.
The move from agentic AI in retail to banking is therefore more than a change of task.
In retail, ecommerce AI agents might compare sellers.
While an AI agent for ecommerce completes a checkout.
AI agents for retail could help explain why customers value automation, but depositing savings calls for clear authority and oversight.
As agentic shopping spreads, the practical question is how much control people hand over.
The appeal of AI shopping agents is that they can act on preferences without repeated effort.
In agentic AI in retail, that might save time; in banking, similar instructions could affect many institutions together.
For now, agentic payments remain a potential channel for the run Slok described. The challenge is to let customers seek better returns while ensuring banks can withstand a faster flow of deposits if automated transfers become common.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Fintech sections to stay informed and up-to-date with our daily articles.
