Bank of America, Morgan Stanley, JPMorgan and Citi are limiting third-party AI agents’ access to customer accounts. According to an October 22 report by Axios, some agents acted in ways that seemed independent of customers’ instructions.
Customers are the ones who give AI agents their banking login details, expecting to delegate financial tasks to them. But a user’s permission doesn’t mean the banks approve of such access: usually, they want customers to log in themselves.
On the surface, banks are worried about errors and unauthorized transactions, but the dispute has an economic angle as well.
Who Is Liable for an AI Mistake?
Axios reports that some agents contacted bank call centers, and in a few cases their actions appeared to exceed the customer’s instructions.
If an agent makes an error when transferring funds or makes a transaction without being instructed by the customer, who pays for the damage? For now, there are no universal rules that clearly assign liability in all such cases. Agreements with a bank may include specific terms, but they don’t automatically resolve every legal question.
Read Also: Coinbase AgentKit: How AI Agents Got Wallets and Started Spending Money on Their Own
Until a clear legal framework with clearly defined lines of liability is in place, it’s premature to call financial agents ready for mass adoption. If money ends up in the wrong place, it isn’t enough to show that a program made the transaction. It also has to be determined who gave it access, who authorized the action and who must cover the loss.
Is It Just About Security?
The banks’ concerns are understandable, but the restrictions have a flip side: they benefit the banks themselves.
An autonomous AI agent can compare rates and fees, find a better product and suggest that the customer move money to another bank. As a result, a financial institution risks losing deposits and control over its relationship with the customer.
For example, JPMorgan is developing its own cash management tool, Smart Cash. It lets the bank offer automation while keeping control over the tool and the conditions it operates under.
JPMorgan CEO Jamie Dimon said the bank doesn’t plan to let just any third-party AI agent into its systems.
“It’s too risky for both our customers and our company,” he said.
Charles Schwab CEO Rick Wurster has also said his firm blocks such programs.
Financial institutions are protecting not only their customers’ money but also their own core business, because agents can find better uses for funds than any given bank representative does. And naturally, banks will guard their customers’ money against being moved to competitors or into other deposits, options and strategies.
Under the guise of security, financial institutions could confine AI assistants to their own ecosystems. Agents could analyze accounts and select products, but probably couldn’t move money to other banks, even when terms there are better. Ultimately, these protective measures may reduce risks while also limiting their own customers’ freedom to choose financial services.
