AI agents bring new risks for banks as the industry races to adopt them

[EventHorizon Crossing]

Politico has a piece on the new risks AI agents are creating for banks, and it lands in a week when agent security has been in the headlines almost every day. Banks are keen to use agents for everything from customer service and complaints to managing savings and payments. The attraction is obvious, with lower costs and faster service. But agents that can act on their own also create problems that traditional banking safeguards were never designed for

The banks themselves have been raising concerns. Last week a group including NatWest and Bank of America warned that AI shopping agents are outpacing consumer fraud protections, and set out principles for what they called trusted AI commerce. If an agent buys something on your behalf and gets it wrong, or is tricked by a scammer, it is not clear who is responsible or how you get your money back

There are wider systemic worries too. Regulators have pointed out that if lots of customer agents react to the same news at once, they could move deposits between banks far faster than humans ever would. That could turn a wobble into a bank run in hours rather than days. Apollo's chief economist Torsten Slok has also warned that agents hunting for the best rates could quietly drain the cheap deposits banks rely on

Then there is the security side we have been discussing all month. Agents can be manipulated with prompt injections, can overstep their permissions and can make confident mistakes. In most industries that is embarrassing. In banking, it means real money going to the wrong place

I can see banks using agents for internal work fairly safely, but letting them loose on customers' money feels like a big step. Would you let an AI agent move money or make purchases for you? And who should be liable when it goes wrong?

Ai Agents Banks New Risks 01096173www.politico.com