AI is logging into bank accounts – and banks are struggling to tell who’s in control

It may not be the case for the average person with a savings account and nothing more, but other customers who have handed their online banking password to an AI assistant are breaking the rules at most banks, and some of those assistants have reportedly done things nobody asked for.

Now America’s biggest banks are shutting them out, while those in Spain are running their own tightly controlled tests.

Banks have found AI assistants logged in as their customers 

Banks across the US have detected external AI agents (software systems that carry out tasks online for you) operating inside customer accounts, according to a report published on Thursday October 22 by Axios. The agents often got in straight through the front door, by logging in with the customer’s own username and password.

That’s against the rules at most banks, which require actual ‘people’ to log in to their own accounts. However, some agents also reportedly did things that looked independent of what the customer had asked, while others contacted bank call centres. Banks say they’re finding it hard to tell whether a customer or an AI started an action, and the worry is greatest for agents that can buy things or move money.

Morgan Stanley has reportedly begun blocking third-party agents, although a spokesperson said the bank hadn’t seen “rogue AI agents accessing our platforms”. JPMorgan, Bank of America and Citi are said to be restricting them too, with Wells Fargo among the banks working to contain the risk. Most banks are allowing, or moving towards, read-only access, which lets a tool view balances but not make payments. 

However, the report relies on anonymous sources, with no figures on how many accounts were involved exactly and no AI tools named, so the scale is still unclear.

JPMorgan’s boss fears ‘thousands’ of rogue agents

JPMorgan, America’s biggest bank, had drawn its line nine days prior. As it reported third-quarter results on Tuesday October 13, the bank confirmed that outside AI agents can’t operate retail customer accounts. They can see some account data on a read-only basis, but they can’t send money or pay bills.

Chief executive Jamie Dimon told the earnings call it was too risky to let “just any outside agent” in, adding: “You don’t want rogue agents, and it could be thousands of them.” A spokeswoman said third parties logging in with customers’ banking details isn’t safe, and customers shouldn’t share their usernames and passwords with anyone.

Regulators want to know how banks would switch an agent off

Regulators are asking the same question from the other side. Reuters reported on Tuesday October 13 that the US Federal Reserve, the Office of the Comptroller of the Currency (a federal banking regulator) and the Federal Deposit Insurance Corporation have been quizzing banks about how they control increasingly autonomous AI.

They want to know what limits apply, how an agent’s actions are monitored and audited, and what stops it going beyond the authority it’s been given. Banks are also being asked how they would shut such a system down, though no formal requirements have been issued yet.

Europe’s supervisors are watching too. The European Central Bank gave banks an end-of-October deadline to produce plans against AI-powered cyberattacks, which is aimed at hacking rather than customers’ own AI assistants. In July, Bank of England deputy governor Sarah Breeden said that as AI agents start making payments for consumers, regulators will have to deal with consent, liability and fraud.

Spanish banks are testing AI payments, but with strict limits

None of this means banks are completely against AI handling your money. Spain’s biggest names are testing just that, but on very different terms. Santander and Mastercard announced on Monday March 2 what they called the first real end-to-end payment made by an AI agent in Europe, run in a controlled setting. The agent worked within limits set in advance, and Santander stressed it was a pilot and not a commercial launch. 

CaixaBank followed with Visa on Thursday July 2, completing a purchase started by an AI agent on a cardholder’s behalf using real card data. The bank said it relied on tokenisation (swapping card details for a coded stand-in), identity checks and fraud monitoring, with the agent working under authorisation set by the customer.

Santander’s test used Mastercard’s Agent Pay system, which creates a unique encrypted token for each transaction so the customer’s real card details are never handed over. ING in the Netherlands and Nordea in Finland also ran similar tests in June. These tests, however, only used single-use passes for one purchase, whereas sharing a banking password hands the AI the keys to the entire account.

Spaniards trust AI with their cards more than Brits do

A survey of 1,750 consumers in Spain, Italy, France, Germany and the UK, run by payments firm Ecommpay and reported in July, found 62 per cent of Spaniards would trust an AI agent to hold their card details. In the UK the figure was 33 per cent.

An important caveat is that the survey comes from a payments company, and measures attitudes towards the concept of letting an AI use your card rather than being based on actual buying experience and real life testing.

Who pays if an AI assistant empties your account?

In Spain, banks must refund an unauthorised payment immediately unless they can prove the customer authorised it or acted with gross negligence, and the bank carries the burden of proof, according to consumer-law explainers. The risk for customers is that those guides list voluntarily handing over your banking credentials as an example of gross negligence, which can let the bank off the hook. Whether giving a password to an AI assistant would count as exactly that for Spanish banks at the moment is untested territory.

The wider rules are behind too. PSD3, the EU’s update to its payments rules, isn’t expected to apply before 2027 at the earliest, and as written it assumes a ‘person’ starts each payment.

Read-only access is the safe route for now

So until Spanish and other banks set out their own rules, the general advice mirrors that which JPMorgan is telling customers. Don’t give an AI assistant, browser tool or any app your online banking username and password. If you want an AI tool to help track spending, look for a read-only connection that your own bank offers.

If you’ve already shared your details with an AI tool, change your password and let your bank know. And check your bank’s terms, because most banks require customers to log in to their own accounts, and ignoring that could leave you arguing with your bank over a refund if something goes wrong.