HSBC, Barclays, Lloyds, NatWest and Nationwide have moved to keep AI agents out of customer accounts, drawing a boundary between digital tools that offer financial advice and those that can act on it. In some cases, the limits ban an agent from moving money without the customer’s direct involvement.
The measures are precautionary. There is no evidence of AI agents accessing UK bank accounts, and the banks have not issued detailed public statements setting out how their limits will work. The central question is who bears responsibility if an automated financial assistant mishandles a customer’s money.
For customers in London and across the UK, the distinction matters because an assistant able to recommend a better savings account is not the same as one authorised to move funds. The banks’ approach places a limit on automated action while regulators develop rules for this emerging form of financial service.
Advice versus action
AI agents are designed to carry out tasks on a user’s behalf, rather than simply answer questions. In finance, that creates a practical divide between helping someone understand their account and exercising authority over it. A recommendation can leave the final decision with the customer. An agent able to act introduces questions about permission, control and responsibility.
The concern is not an established case of wrongdoing by an AI provider or a bank. Lloyds finance chief Will Chalmers said the bank had seen no sign of agents in accounts but was guarding against the risk. Standard Chartered chief financial officer Manus Costello said such tools needed to be safe both for customers and for the bank.
That emphasis on safety reflects the difficulty of assigning liability when a customer gives software a task involving money. The customer’s instruction, the authority granted to the agent and the bank’s obligations all matter. Those issues become more consequential when a tool can execute a decision rather than merely propose one.
The available detail does not establish a common policy across all the named banks. The limits should therefore not be read as identical bans on every use of AI in personal banking. What they do establish is resistance to agents acting inside accounts without the level of customer involvement banks require.
Deposits in focus
The potential effect extends beyond individual payments. McKinsey estimates that up to £700bn of deposits could move from low-paying accounts on AI advice. That is an estimate of possible movement, not evidence that customers have already moved those sums or that agents have gained access to their accounts.
An assistant that helps a saver compare accounts could make it easier to identify a better return. Whether the customer then moves the money personally or allows a tool to do so is a separate question. The distinction is important: advice can influence where deposits are held even when automated account activity is restricted.
McKinsey’s estimate illustrates the possible scale of the commercial stakes for banks. Deposits are central to their business, while savers have an interest in the return their money earns. Neither point establishes that the new limits are intended to stop customers seeking better rates. The stated concern is the safety and accountability of agent-led activity.
Rules taking shape
The UK measures have precedents in the United States, where JPMorgan, Citigroup, Bank of America, Wells Fargo and Charles Schwab have imposed restrictions. That places the British response within a wider debate about how much authority financial institutions should allow automated assistants to exercise.
The FCA is building a framework covering agent identity, delegated authority and accountability. In practical terms, those subjects concern identifying the tool seeking access, defining what a customer has permitted it to do and establishing responsibility for its actions.
The issue is becoming more immediate as Meta’s Muse assistant, which can shop and pay bills, is expected to launch in the UK within weeks. Its expected arrival does not establish that it will have access to bank accounts. The next question is how banks’ limits and the FCA’s developing framework will define what such assistants can do, and where customers must remain directly involved.
