Citi and Bank of America join JPMorgan in restricting outside AI agents' access to customer accounts

Citigroup has joined JPMorgan Chase in barring outside AI agents from logging into customer accounts or moving money, and Bank of America has restricted their access to consumer deposits, as banks weigh security, liability and the risk that cheap deposits leave.

What each bank has said

Citigroup said it does not permit "third-party AI agents or assistants to log into a customer's account or transact directly within our digital banking platforms," according to a statement first reported by the Financial Times on Friday, Oct. 16. The bank said the rule is meant to protect customer accounts and information, and that customers can still let certain open-banking providers see account information on a limited, read-only basis.

Bank of America's chief executive, Brian Moynihan, said in a CNBC interview on Wednesday, Oct. 14, that the bank wants a person to authorize any payment out of an account. He said the bank is making that case to the companies building AI agents, at least until "further development of the capabilities" of agents reduces the risk of errors. The same Oct. 16 report said Bank of America has restricted agents' access to consumer deposit accounts while it works out who would be responsible for payments an agent approves.

JPMorgan Chase moved first. On the bank's third-quarter earnings call on Tuesday, Oct. 13, Chief Executive Jamie Dimon told analysts that the bank would not let just any outside agent operate inside its systems because the risk to clients and to the bank was too high. "You don't want rogue agents," he said. A JPMorgan spokeswoman said the bank does not allow third parties to get into accounts by obtaining customers' banking credentials and logging in for them. "We support secure, customer-authorized data sharing through open banking providers," she said.

What agents can still do

The restrictions stop agents from acting, not from looking. At JPMorgan, an AI tool connected through a data aggregator such as Plaid can still read some information, including balances, but cannot send money or pay a bill. Those are the tasks that would make an agent most useful for managing money. The bank has not given a timeline for letting agents act directly in accounts.

Dimon also put customers on notice. People who use outside agents "should understand there's a liability transfer," he said on the call. In other words, a customer who hands an agent their login may bear the cost if it does something they did not intend.

Why banks are drawing the line now

The moves follow the September launch of Meta's Muse, a personal AI assistant that can connect to users' financial accounts through Plaid and offer guidance on balances, transactions and investment holdings. Bank stocks fell in September as Muse climbed the app download charts, and OpenAI released its own agent, Dots, a few weeks later.

Much of the worry is about deposits. Apollo chief economist Torsten Slok has argued that agents could automatically sweep household cash from bank accounts paying 0.1% on average into fintech accounts paying 3.3% to 5%. Deposits that pay little or no interest are among the cheapest money banks have to lend.

Not everyone expects a sudden exodus. Jeff McMillan, a former head of firm-wide AI at Morgan Stanley who now runs an advisory firm, told American Banker that what Slok describes is "repricing, not a run," and that an agent-driven bank run, if it happens, is "a long way off."

Meta declined to say whether Muse could keep offering financial help if more banks limit access. It said it is developing a personal agent protocol with Sierra, a company that makes AI customer-service software, to make clear to businesses when an agent is acting for a customer.

Regulators are also stressing human approval

The banks' stance comes as international regulators push for human checks on AI agents. On Thursday, Oct. 15, the Financial Stability Board, a global body that coordinates the work of national financial authorities, published its final sound practices for financial institutions adopting AI. The guidance covers firms' own use of AI rather than customers' outside assistants. It calls for human oversight matched to a use case's risk and autonomy, and for "clear, risk-based triggers for when an agent must seek human approval." The board also warned that agents could take unauthorized or mistaken actions.

What comes next

None of the three banks has said what safeguards would let outside agents move money. Dimon said JPMorgan is building tools to protect customers, and Moynihan tied any change to agents becoming more capable. Until then, at Citigroup and JPMorgan, AI assistants can mostly read customers' bank data, not act on it.