Could AI agents speed up runs on individual weak banks?

Within the same rebuttal of Apollo economist Torsten Slok's bank-run warning, economist Cullen Roche — quoted at length by newsletter author Anthony Pompliano — concedes a version of the risk he otherwise dismisses: "the stronger version of this argument deserves a fair hearing. SVB was a run from one bank to other banks, and it still killed the bank. If AI agents can move uninsured deposits in hours instead of days, individual bank runs could happen faster than they used to." He adds that "sticky, low rate deposits have always been part of how banks manage interest rate risk. If that stickiness goes away, weaker banks have to manage their balance sheets more carefully."

The mechanism is speed, not scale. Roche's broader argument (see Would AI cash-sweeping agents trigger a systemic bank run?) is that deposits leaving one bank land at another, so the banking system's total deposits are unchanged. But Silicon Valley Bank's collapse already showed that a single institution can fail even when the money it loses simply reappears elsewhere — the failure is about that bank's ability to meet withdrawals, not about aggregate deposits. Roche's claim is that agents acting on behalf of depositors compress the timeline of that single-bank dynamic, converting what used to take days of coordinated withdrawal into something that could happen in hours, which matters most for banks that already rely on deposit stickiness to manage interest-rate risk. His conclusion: "that's a legitimate risk for specific banks with bad balance sheets. It is not a system-wide run."

This is a distinct mechanism from the general delegation question raised in Can competitive pressure force unsafe agentic AI adoption? — there the risk is firms cutting corners under competitive pressure; here the risk is that giving many individual depositors a fast, automated exit option changes the risk profile of specific banks regardless of how carefully any single agent is built or vetted. It also sits against How willing are people to let AI control their finances?: current reluctance to grant agents banking access is exactly what keeps this risk latent, so the risk Roche describes is conditional on the trust gap closing.

The excerpt gives no figures on how many uninsured deposits are agent-reachable today, no threshold for how much deposit velocity a bank's liquidity buffers can absorb, and no list of which banks already carry the "bad balance sheets" Roche flags. The claim is a economist's reasoned concession in a newsletter rebuttal, not a modeled estimate — the honest implication is that faster-moving deposits raise the bar for bank liquidity management at the institution level, not that a sector-wide early-warning system currently exists to catch it.