The Fed's Barr Warns AI Agents Could Turn the Next Bank Run Into a Matter of Minutes. Banks Are Already Catching Agents Trying to Act in Customer Accounts.
Monday, November 16, 2026. Federal Reserve Governor Michael Barr told a conference in Milan that autonomous AI agents “could instantaneously detect yield differentials or stability signals and autonomously move funds at high speed,” and that bank runs could “unfold in minutes, overwhelming traditional liquidity management and regulatory intervention timelines.”
It is a warning, not a proposal. But it carries one present-tense fact: major banks, Barr said, “now report finding AI agents attempting to take actions in retail customer accounts.”
- Minuteshow fast a run could unfoldBarr's phrase for agent-driven deposit flight; the speech text is on federalreserve.gov.
- $42Bwithdrawal attempts in one daySilicon Valley Bank, March 9, 2023, with about $100 billion more queued for the next morning, per the Fed's own review.
- $800B+debt at the seven largest AI firmsExpected by year-end, Barr said, citing the Fed's Financial Stability Report.
- $3T+off-balance-sheet AI liabilitiesPotential maximum nominal value, per the same report as cited by Barr.
Barr spoke at the Seventh Annual ACM International Conference on AI in Finance. His text makes three separate claims. First, that agents “may accelerate deposit outflows during stress periods” by reacting to rate gaps or stability signals faster than any human depositor. Second, that banks cannot always verify “authenticity, consent, and alignment with customer instructions” when an agent acts in an account. Third, that markets built on a small set of similar models face “herding behavior,” because “a market with large numbers of similar models that may have the same flaw risks failures with systemic consequences.”
His remedies are operational, not statutory: “strict authentication, explicit consent verification for specific actions, technical controls limiting agent capabilities,” and AI-driven decisions that are “well controlled, numerically and legally precise, explainable, and replicable.” Barr described agents attempting actions in accounts. Nothing in the speech or the coverage we reviewed documents an AI-driven run, or a dollar loss, at any bank.
The last big test of how fast a modern run moves was Silicon Valley Bank. The Fed's April 2023 review found depositors tried to pull $42 billion on Thursday, March 9, 2023, with about $100 billion more queued for Friday, a pace the bank's liquidity planning had not anticipated. It failed that Friday, and the review criticized how Fed supervisors had handled the bank.
Barr led that review as the Fed's Vice Chair for Supervision, a post he held until stepping down from it in early 2025; he remains a governor. The 2023 run unfolded over roughly a day, driven by humans on phones. His 2026 argument is that an agent instructed to protect a customer's balance removes the hesitation, and that thousands of agents running similar logic would act together. In February 2025 he made the related point that agents told to maximize profit could converge on the same strategies, per Treasury & Risk.
The same speech turned to the other fragility. Citing the Fed's November Financial Stability Report, Barr said debt at the seven largest AI firms “has been growing significantly in 2026” and is expected to top $800 billion by year-end. Off-balance-sheet liabilities, including data-center leases and energy contracts, now exceed $3 trillion in potential maximum nominal value, and hyperscalers and chipmakers have put more than $100 billion into the labs and cloud start-ups that buy their capacity.
Large U.S. banks, Barr noted, hold substantial capital, so the exposure runs mainly through leveraged hedge funds, where a sharp drop in AI valuations could force fast selling and strain prime brokers. The Fed's May 2026 report found half of surveyed market participants named AI a salient risk, up from 30 percent in fall 2025, per the Fed's spring report.
Barr was nominated to the Board by President Joe Biden (D) and has devoted multiple speeches to AI. We found no Fed rule proposal, guidance, or stress test tied to the agent-run scenario. Bloomberg Television's headline on FDIC Chair Travis Hill, a Trump appointee, below, says the agency will not rule out agentic-AI bank-run risk. That is a statement about uncertainty, not a policy.
Documented: Barr's Nov. 16 speech and its quoted warnings; the SVB run figures in the Fed's 2023 review; Barr's statement that banks are finding agents in retail accounts.
Cited, not independently verified: the $800 billion, $3 trillion and $100 billion AI-finance figures, which Barr attributes to the Fed's November report.
Not shown: any agent-driven deposit outflow, any bank loss, or any regulatory action specific to agents.
The man who reviewed the 2023 bank run now says the next one could happen faster than regulators can react. The evidence so far is agents probing accounts, not draining them, and the response so far is a speech.



