Banks Slam the Brakes on AI Agents They Cannot Trust

U.S. banks are restricting third-party AI agents after detecting account activity they cannot reliably tie to customer instructions. The industry sees useful automation ahead, but says the security and liability gaps are still too wide.
Banks Slam the Brakes on AI Agents They Cannot Trust

Banks Slam the Brakes on AI Agents They Cannot Trust
The first alarm is not that AI agents can enter a bank account. It is that banks often cannot tell whether they are acting for the customer — or for themselves.

Major U.S. lenders, including Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and Citi, have detected third-party agents inside customer accounts, Axios reported. In some cases, customers had supplied credentials; in others, banks tracked agents contacting call centers or appearing to act without clear instructions. One executive said the industry struggles to determine whether an agent is authorized, has gone rogue or has obtained access improperly.

That uncertainty has driven a fast retrenchment. Morgan Stanley has begun blocking third-party agents, while JPMorgan, Bank of America and Citi are restricting them. The banks’ concern is straightforward: an agent with permission to purchase or move funds could do far more damage than a read-only budgeting app. Yet lenders also acknowledge customers will want these tools, leaving them to search for a middle ground built around limited access and stronger controls.

The problem has quickly become a wider industry headache. A separate account summarized the emerging reality bluntly: banks have found agents accessing accounts and taking actions “sometimes without clear authorization or instruction.”

Warnings about the technology had already been mounting. At an Axios event on Oct. 20, FDIC Chairman Travis Hill said the crucial question was how much financial decision-making depositors would delegate to agents. Alpaca CEO Yoshi Yokokawa raised the harder follow-up: “Who is liable when something happens, when the agents go rogue and start doing some criminal activity?”

J.P. Morgan payments and global banking data chief Zack Anderson captured the contradiction facing lenders: getting an agent to make a payment is easy; doing it “reliably and safe” is not. For now, banks are choosing caution over convenience.

https://foxvector.com

Write a comment