AI Agents and Banking Relationships

Banking, Agents, and Relationships

Banks, doctors and psychiatrists have long assumed that their customers and clients would always prize human skill, qualifications and experience over advice offered by free-to-consumer chatbots. But it turns out that consumers are quite willing to look to chatbots such as ChatGPT for advice on finances, physical and mental health, and relationships, with many people finding a chatbot easier to talk to than a human. A chatbot won’t scold you for spending money on coffee or avocados, and on the surface, all of this advice appears to be free.

So it won’t be a surprise if consumers adopt new AI agents that offer to act as digital personal financial advisors that do your bidding by making the most out of your current account – and suggesting you move money to interest-bearing accounts. Or, God forbid, telling you to invest in cryptocurrencies, or put down a bet on the next election.

AI agents are not well understood, but you can think of them as AI tools with agency. Rather than waiting for instructions, they will work away in the background without needing constant instructions from users. Do bear in mind that an “AI Agent” is also a marketing tool suggesting a great deal of autonomy. Recent stories of AI agents “escaping” their supposedly controlled environments suggest that these agents are powerful, and some people consider them an existential threat to humanity.

What if they are just robots looking for relationships?

The AI agents looking for relationships

John Collison of Stripe runs a regular podcast called Cheeky Pint, and he recently sat down with Mark Zuckerberg and Meta’s Chief AI Officer Alexandr Wang. Meta has just launched its new AI agent Muse, while Stripe has been at the forefront of developing machine payments protocols. It’s worth a watch to see what’s coming in the world of AI agents, if you have two hours free!

View Cheeky Pint podcast

By some estimates, 3.6 billion people use one of Meta’s services (Facebook, WhatsApp or Instagram) on a daily basis, and Meta offers its new Muse service with a significant free tier. Muse was launched on 8 September 2026, quickly rising to top the free app download charts. It promises to do task such as shopping and can make purchases using Stripe’s Link, which pays with a stored payment methods or a one-time virtual card. It asks the customer to approve each purchase but can’t initiate bank-to-bank transfers or move money from a current account to savings.

The reaction of other major tech players – and banks – was swift. Within two weeks, Amazon had blocked Muse from its platform. Amazon had already blocked AI app Perplexity, and other tools including those from OpenAI and Google from acting as intermediaries – and Amazon has Alexa as its own AI-driven agent.

Earlier this month, major US banks followed suit. On an earnings call Jamie Dimon said: “We’re not going to allow just any outside agent to come inside the company”. Charles Schwab CEO Rick Wurster said: "We don't want someone just giving their password to Muse, and letting Muse come in and take action on that client's behalf. We can't control it, the client can't control it, something could go wrong and who's liable?” Brian Moynihan of Bank of America said the bank feels very strongly that there should be a human involved in authorising before money moves out of customer accounts, citing liability, fraud, and unapproved cash movements.

Those are all reasonable concerns but they leave out one crucial point. The minds of bank leaders were focused by a report late in September from Apollo economist Torsten Slok, with Slok warning that AI assistants could automatically sweep household cash into accounts paying 3.3 to 5 per cent instead of the 0.1 per cent national average on checking accounts. “JPMorgan, Bank of America and Wells Fargo hold $1.6tn of non-interest-bearing deposits — 16 per cent of their liabilities,” the FT reported. “Paying a 3 per cent rate on those would cost them $47bn a year, almost half of their collective annual earnings… If every household used AI agents to optimise the return on their cash balances, banks could lose a large share of the cheap deposits they rely on to make loans, which would be a problem for the entire financial system.”

This development has been a long time coming. Few things worry bankers as much as the challenge of gaining and keeping customer deposits. As digital banking and mobile and digital wallets have surged in popularity, many people, especially in emerging markets, immediately move their salaries into digital wallets to do their everyday transactions. Stablecoin operators in the US can’t offer interest bearing accounts, but they are trying to offer interest through exchanges and affiliates, generating fierce opposition from the huge community banking sector.

It’s worth bearing in mind that some banks – such as Revolut and N26 – already offer cash optimisation features, and it’s a common offering in business and SME accounts, to help businesses maximise cash working for them.

Is this a re-run of the screen scraping controversy?

Klarna chief executive Sebastian Siemiatkowski argued recently on X that this new AI agent debate was an update of a much earlier debate about screen-scraping. In the mid-2000s, German business Sofort built a digital interface to enable bank-to-bank payments (as many Germans did not have credit cards) for online transactions. (It didn’t help that German banks had notoriously difficult and unfriendly payments systems). Customers could allow Sofort to access their bank accounts and initiate payments, and banks balked at the development, and lobbied the EU regulators to prevent third-party access.

Banks feared losing their customers to services that were easier to use and had better customer interfaces, which fintechs often provided. In Germany, the Federal Cartel Office ruled that banks barring customers from using their PINs with services such as Sofort was uncompetitive, reinforcing the principle behind PSD2 access rules. (Klarna acquired Sofort in 2014, and eventually integrated it completely into the Klarna offering.)

Customer authentication and authorisation is particularly important

Under the EU's PSD2 rules, banks must meet strong customer authentication and payments must be authorised by the customer. Now regulators are thinking about how to regulate AI agents, and discussing Know Your Agent (KYA) rules.

Is a customer instruction such as “Manage my money” a strong enough authorisation for an AI agent to go ahead and transfer customer money? How does a bank verify that the customer approved the payment, and who is liable if the agent sends money to the wrong account? While we’ve yet to fully solve the question of digital identity for people, we’ve probably now come to the point where ‘I’m Not A Robot’ tests are no longer going to be sufficient.

For the time being, the focus of regulators is to keep humans in the loop, and for the human account owner to approve any transfer of money suggested or initiated by an AI agent.

The threat to deposits and relationships

Banks are realising that AI agents could fundamentally weaken their control over customer relationships and deposit balances. New apps including Bank AI and Herbert are promising just these features. Herbert, for instance, connects your accounts through Plaid, monitors your money and then texts you with suggestions such as moving money out of zero or low interest accounts.

In the US, the big players are offering AI financial tools, and converging on the model which uses Plaid to connect your accounts so they can read your financial data across different banks, cards, and wallets. For now, they’re mostly positioning their AI agents as analysts and advisors.

The AI Agent doesn’t have to make transactions to win your relationship

For regular bank accounts, the customer makes the decision to use or even check their bank account. With AI tools, your agent may be constantly monitoring and suggesting ways to maximise your balances. If this works, soon you’re interacting more with the AI agent than you are with the bank, and the relationship shifts.

So the concerns of bank bosses about AI agents going rogue and transferring money is not (yet) the real worry. Just as Sofort threatened the German banking model, AI agents could become the intermediaries between customers and their bank accounts. With Sofort and fintechs, banks fought third-party access, lost, and open banking followed.

Who will bet against Muse and its agent competitors this time around?