Could AI Trigger the Next Bank Run? Federal Reserve Warns of New Financial Risk
Artificial intelligence (AI) agents could trigger bank runs in minutes by automatically moving customer deposits in response to interest-rate changes or financial instability signals, Federal Reserve Governor Michael Barr has warned.
The warning highlights a new financial stability risk as autonomous AI tools enter consumer banking, while mounting AI infrastructure debt and complex financing arrangements add pressure to the financial system.
AI Agents Could Accelerate Deposit Withdrawals
Speaking at an Association for Computing Machinery conference in Milan, Italy, Barr warned that autonomous AI agents could fundamentally change how quickly customers withdraw money during periods of financial stress.
Unlike traditional bank runs, which require customers to initiate transfers individually, AI agents could monitor interest-rate differences and signals of financial instability before moving funds automatically.
“Where customers once needed to execute transfers, AI agents could instantaneously detect yield differentials or stability signals and autonomously move funds at high speed, potentially accelerating runs that unfold in minutes, overwhelming traditional liquidity management and regulatory intervention timelines,” Barr said in prepared remarks, according to Bloomberg.
The risks extend beyond automated withdrawals. Barr warned that banks could also face AI agents attempting to access customer accounts or perform transactions without proper authorisation.
Banks would then need to determine whether such requests were legitimate and supported by customer consent, potentially overwhelming existing verification systems.
Banks have already identified agents attempting to access retail customer accounts without authorisation, according to an earlier Axios report.
Barr called for stronger authentication protocols, explicit consent verification and technical safeguards to restrict what AI agents can do.
He also stressed that AI-driven decisions in consumer credit and investment advice must remain “well controlled, numerically and legally precise, explainable, and replicable.”
AI Infrastructure Debt Raises Concerns
Barr’s concerns also extend to the financial arrangements supporting the AI boom. He highlighted more than $3 trillion in potential off-balance-sheet obligations linked to AI infrastructure spending, including data centre leases, energy supply contracts and collateral guarantees.
According to the Federal Reserve’s latest Financial Stability Report, on-balance-sheet debt among the seven largest AI companies is expected to exceed $800 billion before the end of the year.
Another concern is circular financing, in which major technology companies and chipmakers invest in smaller AI startups that subsequently use the money to purchase computing capacity from those same industry players.
Such arrangements could amplify financial stress if returns from AI investments fall short of expectations.
Barr also warned that elevated leverage among hedge funds and equity strategies could trigger rapid deleveraging and counterparty pressure if AI valuations experience a sharp correction, despite substantial capital buffers at large U.S. banks.
Insurers and AI Trading
The growing financial exposure to AI infrastructure is also reaching the insurance industry. A New York Times report found that U.S. life insurers Athene and Corebridge Financial held $85.2 billion in AI-related debt as of June, citing Moody’s data.
Barclays separately estimates that life insurers own around 15% of all AI data centre debt. Moody’s projects these investments could reach $185 billion by the end of 2027.
Some financing arrangements have also raised concerns about how AI-related borrowing is disclosed. The New York Times reported that xAI used a separate financial entity to borrow money to purchase chips and lease them to the company, an arrangement intended to keep the borrowing off its books.
SpaceX’s auditors ultimately determined that its borrowing through a similar arrangement needed to be recorded as company debt. However, another $40 billion in financing has reportedly been arranged through a separate entity.
Barr also cautioned that widespread dependence on similar AI trading models trained on overlapping datasets could encourage synchronised trading behaviour and amplify market volatility.
Cybersecurity presents another challenge. Autonomous AI models may treat security restrictions as obstacles rather than rules, increasing the risks facing financial institutions as AI-powered attacks become more sophisticated.