In a speech at the Association for Computing Machinery’s International Conference on Artificial Intelligence in Finance in Milan, Federal Reserve Board Governor Michael S. Barr highlighted how financing the artificial intelligence investment boom and expanding use of the technology could amplify financial stability risks. The assessment emphasized monitoring complex financing exposures and controlling risks in trading, cybersecurity and consumer finance.
The speech reviewed findings from the latest Financial Stability Report, including expectations that debt on the balance sheets of the seven largest AI firms will exceed USD 800 billion by the end of 2026. Other obligations, including contracts, leases and guarantees, have a potential maximum nominal value exceeding USD 3 trillion. These obligations are generally reported before discounting, complicating comparisons with balance sheet debt. Circular financing and collateral guarantees connect larger technology firms with weaker borrowers and can obscure where losses would fall. Although Barr expressed less concern about direct losses given the still limited debt backing the boom, a sharp repricing could trigger rapid deleveraging and fire sales among leveraged investors, increasing banks’ counterparty exposures.
In trading, reliance on similar AI models could synchronize market reactions, while some learning algorithms could develop collusive or manipulative strategies without explicit instructions. AI also increases cyberattack capabilities, including exploitation of vulnerabilities in shared service providers. For consumer finance, Barr called for explainable models, continuous fairness monitoring, detailed validation records and accessible dispute mechanisms to address opaque lending decisions and investment recommendations that may embed conflicts of interest. Banks have reported AI agents attempting actions in customer accounts without verifiable authorization, underscoring the need for strict authentication, consent checks for specific actions and limits on agents’ capabilities. Autonomous transfers could also accelerate deposit runs beyond traditional liquidity management and regulatory intervention timelines.