Banks push back on AI agents in accounts

What happened

"Banks" was among the day's trending searches after Axios reported on 22 October that major US banks are detecting unvetted third-party AI agents operating inside customer accounts. According to the report, which cites people familiar with the matter, some agents are attempting actions that appear to be independent of user instructions. Banks named in coverage of the report include Morgan Stanley, JPMorgan and Bank of America.

How the agents get in

In most cases the agent is not breaking in. The customer hands it their own username and password, and the agent logs in as if it were the customer. The Silicon Review, summarising Axios, notes that this breaks the terms of service of nearly every major bank, which require people to log in to their own accounts themselves. It also means a bank struggles to tell whether an agent is following the customer's wishes or has gone off script. Engadget reports that some agents contacted bank call centres without specific permission.

What banks are doing

Morgan Stanley has begun blocking third-party AI agents from its platforms, according to the reports, although a spokesperson said the bank has not observed "rogue AI agents" accessing them. A Chase spokesperson told Axios that customers should not share their usernames and passwords with anyone. Many banks are moving towards read-only access, which lets an agent view account information without taking action, similar to the access budgeting apps already get. The Silicon Review adds that banks are also planning "day after" scenarios to restore customer access if an agent causes damage. JPMorgan chief executive Jamie Dimon warned on the bank's 13 October earnings call that there could be thousands of rogue agents operating across financial platforms.

Who pays if something goes wrong?

Liability is the open question. Engadget writes that, under the US Electronic Fund Transfer Act, banks would likely not have to refund a customer who voluntarily gave an agent access, and TFTC reports that Dimon described this as a "liability transfer" to the customer. TFTC also notes that, at the time of writing, no US banking regulator had published a rule specifically on AI agents authorising payments. That is the coverage's reading of the law, not legal advice.

Why it matters for payments

AI assistants are moving from answering questions to taking actions, and moving money is the most sensitive action of all. An agent that can log in with your password can also start a payment or an international money transfer, and a bank payment you have approved usually cannot be cancelled. Open banking takes a different route: an app connects to your bank only with your consent, and you approve each payment with your own bank instead of handing over a password. Read-only access for agents follows the same idea: let software see the numbers, but keep the decision to move money with the person.

What to watch next

Watch for banks publishing rules for approved agents, for industry standards that let an agent prove who it acts for and what it may do, and for any regulator guidance on liability. Until then, the simplest protection is the one the banks repeat: do not give your banking password or one-time codes to any app or assistant.

What it means for Odix Pay users

When you add money to Odix Pay with open banking, you log in directly with your bank and approve each payment in your banking app, and your banking username, password and codes are never shared with Odix Pay. In the same money transfer app you can send money to bank accounts at home and abroad, with the exchange rate and fee shown before you confirm. Keep your Odix Pay login, one-time codes and card details to yourself, and that includes AI assistants. Availability depends on your country, and fees are shown in the app. More stories in the Odix Pay newsroom.