The Agentic Bank Run: How Your AI Assistant Could Drain Your Bank

Wall Street just issued a warning — and it's not about a crash. It's about your AI assistant.

On September 27, 2026, Torsten Slok, chief economist of Apollo Global Management (a private-equity giant managing roughly $1 trillion in assets), published a short note with an explosive title: "Is an Agentic bank run coming?"

His answer: maybe. And the mechanism is unlike any bank run in history — because there would be no panic. Just optimization.

Your Bank Pays You Almost Nothing

Here is the quiet scandal at the heart of modern banking. The national average interest on a US checking account is a miserable 0.1%. On $10,000, that's $10 a year. Meanwhile, fintechs like Revolut, SoFi, Varo, LendingClub and Wealthfront pay between 3.3% and 5% — up to fifty times more. On that same $10,000, that's up to $500 a year.

Most people never notice the difference. According to McKinsey research, most customers don't even know what their bank pays them. And that ignorance is not a bug in the banking system. It is the banking system.

Banks survive on your inertia. Transaction accounts — the checking accounts Slok is talking about — total $8.3 trillion out of roughly $26 trillion in bank liabilities, and banks pay a paltry ~1% on them. Accounts bearing zero interest make up about 20% of the deposit base. The Financial Times has put the zero-interest deposits at JPMorgan Chase, Bank of America and Wells Fargo alone at about $1.6 trillion. These cheap deposits are the fuel banks use to make loans. Slok's warning: agentic AI is about to remove the inertia wholesale.

AI That Acts, Not Just Answers

The trigger is what the industry calls agentic finance — AI that acts rather than just answers. Meta's personal agent Muse can already read your account data through services like Plaid. JPMorgan is planning its own automatic cash transfers during 2026. Another agent called Instinct has gained traction.

Imagine telling your AI assistant one sentence: "make my cash work as hard as possible." It would monitor your balances in real time, compare returns across institutions, sweep idle money into the highest-yield account — and move it back in time for your bills. Continuously. Without hesitation. It doesn't wait for business hours, it doesn't second-guess, and it doesn't care that everyone else's agent is making the same call.

The plumbing for machine-speed money is already being built. Coinbase's x402 protocol — now an open standard under the Linux Foundation's x402 Foundation, with members including Cloudflare, Google, Visa, Mastercard, AWS, Circle and Stripe — lets AI agents pay for services in stablecoins within seconds, no account, no card, no human approval. It has reportedly processed around 200 million cumulative transactions with roughly 69,000 active agents. Mordor Intelligence puts agentic AI in financial services at $7.78 billion in 2026, projected to reach $43.52 billion by 2031.

The Math That Should Scare Banks

Slok's argument is brutally simple: "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, which would be a problem for the entire financial system."

Morningstar's Steve Goldstein ran the numbers. US banks collectively earned $296 billion last year. Forcing banks to pay real interest on roughly $10 trillion of deposits would cost between $200 and $300 billion. In other words, if banks did nothing, their profits would be all gone.

Of course, banks would do something: raise loan rates, cut back on riskier lending, merge to save costs. They absorbed a similar shock after COVID when rates climbed — Silicon Valley Bank was one casualty. But the 1980s savings-and-loan crisis showed how a funding squeeze can turn ugly.

A Bank Run With No Panic

Here's what makes this genuinely new. A traditional bank run happens when depositors lose confidence and rush to withdraw before the money runs out. An agentic bank run would happen because agents are doing exactly what they were told to do: chase the best available yield, at machine speed and machine scale.

Deposit funding assumes friction. People are slow to move money, they tolerate low interest out of habit, and they rarely reallocate all at once. That stickiness lets banks lend long against short-term deposits. Strip out the friction and the model strains. Millions of agents acting on the same signal at the same moment looks less like a crowd and more like a coordinated drain.

The irony is delicious: the same AI hype that made tech stocks soar could be the thing that breaks the banks' business model. The money isn't running. It's being optimized — by machines that never sleep, never panic, and never forgive a bad rate.

And the question it leaves behind is the one worth sitting with: when your AI moves your money better than you ever could — who's really in charge?