JPMorgan, Citi and Bank of America shut AI agents out of accounts

Citigroup and Bank of America have taken similar steps, the Financial Times reported on Oct. 16. Citi said it does not allow third-party AI agents or assistants to log into customer accounts or carry out transactions on its digital banking platform. Customers can still give some open-banking services limited, read-only access to account data, the FT reported.

Schwab too: Charles Schwab CEO Rick Wurster told Bloomberg the firm is actively blocking unauthorized agents, naming Meta's Muse. "We don't want someone just giving their password to Muse, and letting Muse come in and take action on that client's behalf," he said. "Something could go wrong and who's liable?"

Bank of America CEO Brian Moynihan said the bank is working with agent developers on rules for who is liable when AI starts a payment. "We strongly believe there should be a human in the loop to authorise money moving out of your account," he said, as quoted from the FT report.

The money question: Banks also worry agents could sweep customers' idle cash into higher-paying accounts. Apollo chief economist Torsten Slok estimated that agents could move money from checking accounts paying about 0.1% on average into accounts paying 3.3% to 5%, the FT reported. He said JPMorgan, Bank of America and Wells Fargo hold $1.6 trillion in deposits that pay no interest, and paying 3% on them would cost the three about $47 billion a year, close to half their combined annual profit.

How it started: Meta launched Muse on Sept. 8 as a personal agent that can shop, book travel and, through a link with fintech firm Plaid, look at users' balances and investments. Amazon began blocking Muse from its store on Sept. 20, saying the agent did not identify itself and appeared to capture customer passwords, GeekWire and Bloomberg reported. Meta has said Muse cannot see users' passwords or payment methods.

Meta's answer: Meta declined to tell Fortune whether Muse could still offer money help if more banks shut it out. Instead it pointed to "Poppy," a protocol it is building with AI firm Sierra so an agent can identify itself and show it has the customer's permission.

The strongest case from each side: Banks say an agent holding a customer's password is an uncontrolled third party. If it misfires, no one can say who pays, and fraud is harder to spot. Agent makers and their backers say customers should be free to choose their own tools, and that blocks also protect banks' cheap deposits. Citi CEO Jane Fraser has said agents need strict identity checks before they can transact, which suggests banks object to unverified agents more than to agents in general.

What's still unknown: Whether banks will accept a standard like Meta's Poppy, and on what terms; whether regulators or Congress will write rules on who is liable when an AI-approved payment goes wrong; and whether other big banks, such as Wells Fargo, will follow. The FT reported that Federal Reserve officials privately welcomed the limits, citing one unnamed source; the Fed has not said so publicly.

Citi's and Bank of America's restrictions were first reported by the Financial Times on Oct. 16; the FT article is paywalled, and its details are confirmed here through ET Now's account and the banks' public statements quoted by Fortune, Bloomberg Law and The Verge. The $47 billion figure is an estimate by Apollo chief economist Torsten Slok, as reported by the FT, not a bank disclosure. Reports that Federal Reserve officials privately welcomed the limits rest on an unnamed source and are not confirmed.