NextFin News - Torsten Slok, chief economist at Apollo Global Management, has issued a warning that cuts to the heart of how banks fund themselves: artificial intelligence agents that optimize household cash yields could strip banks of the cheap, sticky deposits they rely on to make loans, in what he calls an "agentic bank run." The warning, delivered in a television interview on Oct. 2, 2026 and laid out in a research note titled "Is an Agentic Bank Run Coming?", lands as the fixed-income market braces for a $14 trillion wave of debt supply over the next year. The two forces together — a behavioral shift in who holds household cash, and a historic supply shock in bonds — mean the industry needs to think harder about fixed income, and fast.
The Situation: Cheap Deposits Meet a Debt Deluge
US households currently hold roughly $5.4 trillion in bank deposits earning as little as 0.1%, Slok said. That idle cash is the quiet foundation of the banking system: it is cheap funding that banks on-lend at a profit. But AI personal assistants such as Meta's Muse could change that by automatically sweeping household balances into accounts paying 3.3% to 5.0%.
"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," Slok said in his Sunday note.
The timing is awkward. The Treasury is tasked with refinancing about $10 trillion in existing debt while funding a roughly $2 trillion budget deficit, and gross investment-grade corporate issuance is on track to hit $2 trillion this year as hyperscalers race to fund AI infrastructure. That is a $14 trillion supply shock hitting a market just as its cheapest source of demand — household cash parked in banks — becomes contestable.
The 10-year Treasury yield was trading at 5.27% as of Oct. 2, 2026, two basis points from its 2007 peak, after a late-September selloff that pushed both the Dow Jones Industrial Average and the S&P 500 to monthly losses. The Federal Reserve's benchmark rate sits at 3.75% to 4.00%, down from a peak around 5.4% held for much of last year, but the long end of the curve has refused to follow short rates lower. Traders priced in a 42% probability of another Fed rate move at the next meeting, a sign that policy expectations are still being repriced rather than settled.
The debt burden behind the supply wave keeps growing: US national debt has crossed $40 trillion, and the 30-year Treasury yield has touched its highest level since 2007, above 5.3%. This is not a cyclical dip in borrowing; it is a structural elevation in the government's demand for capital. Interest costs on legacy debt rolling over at today's rates have already pushed federal interest expenses past $963 billion in the first 10 months of the fiscal year, according to Treasury data.
Why This Is Different From Every Deposit Flight Before
Deposits have left banks before. After the rate-hiking cycle began in 2022, total US bank deposits fell from nearly $18 trillion to a little over $17.5 trillion as households discovered money-market funds. But that flight was manual, slow, and incomplete — it required a human to notice the rate gap, open an account, and move the money. Most cash never moved.
AI agents remove the friction. The difference is not the size of the yield gap; it is the speed and completeness with which the gap gets arbitraged. A household that tolerates a 0.1% checking yield out of inertia will not instruct an autonomous agent to do the same. Once the agent is empowered to act, the "laziness tax" on idle cash disappears by default.
This is the mechanism: deposits are not just liabilities on a bank's balance sheet; they are the margin between what banks earn on loans and what they pay for funding. Lose the cheap deposits, and the funding mix shifts toward wholesale markets — certificates of deposit, brokered deposits, senior unsecured bonds — all of which price off the very yields that are rising. Net interest margins compress. Loan pricing rises. Credit supply tightens. The transmission runs straight from a smartphone setting to the cost of capital for mortgages, autos, and small business.
The 2023 regional-bank stress episode offered a preview of deposit sensitivity, and it took a new Federal Reserve lending facility to stabilize the system. The agentic scenario differs in one critical respect: the 2023 outflows were driven by fear and concentrated among large, rate-insensitive depositors. An agent-driven outflow would be driven by optimization — calm, automated, and available to every household with a balance, not just the wealthy. Fear subsides; arbitrage does not.
There is also a distributional asymmetry worth noting. The households most likely to adopt agentic cash management first are precisely those with the largest balances — the same deposits that are cheapest for banks to keep. A 10% outflow concentrated in the top tier of accounts does more damage to funding costs than a 10% outflow spread evenly, because the marginal dollar replaced comes from wholesale markets at a higher rate. Banks do not lose their average cost of funding; they lose their marginal cost advantage.
Cyclical or Structural? Both, and That Is the Problem
The cleanest way to frame Slok's warning is to separate two forces that are usually analyzed apart.
The cyclical leg is the $14 trillion debt wave. Supply shocks of this size are mean-reverting: issuance windows close, deficits narrow, refinancing bulges pass. History offers precedent — the post-financial-crisis issuance glut, the 2020 pandemic supply surge — and each was eventually absorbed as demand caught up. If this were only a supply story, the playbook would be familiar: duration up front, wait for the cycle to turn.
The structural leg is the agentic deposit shift. This is a regime change in household behavior, not a cycle. Three pieces of evidence support the structural read. First, the driver is technological and permanent: once an agent can move money at zero marginal effort, the old equilibrium — cash sleeping at 0.1% — does not come back on its own. Second, the history that anchored the old model no longer applies: the assumption that households leave cash idle because of inertia is precisely the assumption the technology destroys. Third, there is no self-correcting mechanism: banks cannot pay their way out of this without sacrificing the margin that makes deposit-taking profitable in the first place.
Blend the two and you get the uncomfortable conclusion: even as the cyclical supply shock eventually recedes, the structural repricing of bank funding persists. Fixed income is not just waiting out a cycle; it is adjusting to a new funding architecture.
The Second-Order Trade Nobody Is Pricing
The first-order read of Slok's note is obvious: banks lose cheap funding, margins compress, financials underperform. That is already conventional wisdom and is reflected in the attention markets have paid to deposit competition — a broker note this week flagged JPMorgan, Wells Fargo, Charles Schwab, and Morgan Stanley as the firms most exposed if AI-powered tools make it easier for clients to compare cash yields and advisory fees across platforms. Morgan Stanley shares declined more than 2.5% as the deposit-disintermediation argument circulated.
The second-order implication is sharper and less discussed. The same AI boom that creates the agentic-run risk also creates the demand for the bonds being issued. Hyperscalers are borrowing to build data centers; if AI agents successfully redirect household cash into higher-yielding instruments, that cash becomes the pool that buys the very debt funding the AI buildout. In other words, the agentic bank run is not purely destructive — it is also the recycling mechanism that matches a historic supply wave with a newly mobilized demand base.
That reframes the question. The risk is not that $14 trillion of bonds cannot be placed. The risk is the price at which they clear: higher yields, wider credit spreads, and a term premium that prices the uncertainty of who the marginal buyer is. The term premium — the extra compensation investors demand for holding long-duration risk instead of rolling short-term bills — acts like a fear tax on the long end of the curve. It has been dormant for much of the post-2008 era, when central-bank purchases suppressed it. A structurally more expensive funding regime would wake it up, and a term premium that moves from dormant to persistent is what turns a yield spike into a regime change.
The 10-year yield at 5.27% may not be the peak; it may be the market beginning to price a structurally more expensive funding regime. And the circularity runs deeper: the economy now needs AI to generate the cash flow that justifies the debt, while the debt funds the AI that may disintermediate the banks holding the economy's cash.
"If this doesn't happen, then the risk is that the AI trade weakens, with credit spreads widening, capex plans getting cut and ultimately US GDP growth slowing," Slok wrote in earlier research on the AI spending boom.
Consensus estimates on Wall Street expect hyperscaler operating cash flow to grow from $600 billion in 2025 to roughly $2 trillion in 2030 — more than tripling within five years, per data compiled by Apollo. That cash-flow ramp is the assumption holding the whole structure together: the debt gets issued on the promise that the AI buildout pays for itself. If the agents arrive before the cash flow does, the funding gap widens at exactly the moment banks are losing their cheapest source of capital.
The Counter-Thesis: Adoption Friction Is Real
The strongest argument against Slok's call is practical, not theoretical. Empowering an AI agent to move household money at scale requires regulatory approval, consumer trust, and integration with banking rails — none of which arrives overnight. Money-market yields have already fallen from their 2024 peaks (Vanguard's Federal Money Market Fund yielded 5.23% in 2024, 3.77% in 2025, and under 1% in early 2026), which narrows the incentive to sweep. And banks are not passive: they have already begun paying up for deposits, as the shift from $18 trillion to $17.5 trillion in deposits shows they are fighting for balances.
This counter-thesis is credible on timing. Adoption will be gradual, regulators will intervene before a disorderly run, and the yield gap that motivates sweeping is itself cyclical. But it does not defeat the structural point — it only argues for a slower arrival. A regime shift that takes five years instead of one is still a regime shift, and fixed-income portfolios positioned for a quick reversion will be wrong for five years.
The falsifying signal is specific: if US household deposits at banks turn positive and hold there for two consecutive quarters while AI-agent adoption continues to rise — meaning cash stays in banks even after the technology to move it is widely available — then the agentic-run thesis is wrong. Watch the Federal Reserve's H.8 release and the Investment Company Institute's money-market fund flow data. Deposit share stabilizing or rebounding, not just the level of rates, is the metric that matters.
What To Watch
Short term (liquidity and sentiment): The 10-year yield's ability to hold below 5.5% and the bank sector's reaction to deposit-flow data. A break higher in yields alongside widening financial credit spreads would signal the market is beginning to price the funding-risk channel. The 42% probability traders assign to the next Fed move is the consensus anchor to watch — a rapid shift in that number would ripple through the curve.
Medium term (fundamentals): Quarterly deposit trends at the largest banks and the pace of money-market fund inflows. If deposits stabilize while money-market growth slows, the sweep dynamic is contained. If deposits keep falling as agent adoption rises, Slok's mechanism is activating. The $18 trillion to $17.5 trillion decline since 2022 is the baseline; a reacceleration of outflows would be the tell.
Long term (structural): The regulatory response to agentic financial activity and the term premium embedded in long-duration Treasuries. A persistent elevation in the term premium — the compensation investors demand for holding long-term risk — would confirm that the market views the funding regime as permanently changed. With the 30-year yield above 5.3% and national debt past $40 trillion, the long end is already sending a warning.
Base case: gradual adoption, contained deposit outflows, yields settling in a higher range than the 2010s but below current panic levels. Upside case for yields: rapid agent adoption coincides with continued deficit financing, pushing the 10-year toward 6%. Downside case: regulatory friction stalls agent-enabled sweeping, deposits stabilize, and the long end rallies back toward 4.5% as the supply wave passes.
The takeaway is not that banks are about to fail or that households should move every dollar tomorrow. It is that the assumption of sticky, cheap household deposits — the silent subsidy that made the post-2008 funding model work — is now a contestable assumption. Slok's point is that fixed income can no longer treat household cash as inert. In a world where AI agents optimize every basis point, the deposit base is not a balance-sheet constant; it is an option that households have just been handed, and the industry has not priced it yet.
Explore more exclusive insights at nextfin.ai.
