Michael Barr urges consent checks for AI banking agents
Federal Reserve governor Michael Barr warned that banks cannot always verify who authorised artificial intelligence agents trying to act on customer accounts. He called for explicit consent, strong authentication and limited powers. His Milan speech also examined credit discrimination and the danger of common trading models producing similar decisions, while stressing that he was expressing his own views.
Economics & Markets··Morning
Agents raise questions about customer permission
Michael Barr, a governor of the US Federal Reserve, examined artificial intelligence risks in financial services during a November 16 speech in Milan, Italy.[1], [2]
He said major banks had encountered AI agents attempting actions on retail customer accounts. Banks could not always verify authenticity, customer consent or alignment with instructions. Barr called for strict authentication, explicit permission for particular actions and technical limits on agents’ capabilities.[1]
Alternative credit data can carry discrimination risks
Rental payments, utility bills and employment information could widen access to credit, Barr said. But such data could also act as proxies for protected characteristics including race or national origin. Consumers might lack the inspection and dispute rights available for traditional credit reports. He advocated explainable models, continuing fairness monitoring, model validation and accessible procedures for challenging decisions.[1]
Common models could synchronise market decisions
In wholesale markets, Barr discussed the danger of similar models producing similar trading decisions and failing together. He described algorithmic collusion as a theoretical risk rather than an established market outcome. Agents could also accelerate deposit withdrawals during stress, while AI tools increase the capabilities of both cyberattackers and defenders. Barr stressed that these were his own views rather than the Federal Reserve’s position.[1]