Five major UK lenders — HSBC, Barclays, Lloyds, NatWest and Nationwide — have moved to block or restrict third-party AI agents from accessing customer accounts, according to a Cointelegraph report on 16 November. The restrictions are precautionary: no agent access to UK accounts has yet been documented, but banks are concerned about liability if an automated system mishandles a customer's money.
The immediate worry is accountability. Banks' existing dispute processes were built for customers making a few payments a week, not for agents making thousands of small transactions. Edgars Nemše, chief executive of the GenLayer Foundation, told Cointelegraph that even a small share of failures would generate a large number of disputes, and the machinery for handling them still has to be built. There is also a commercial motive: agents could strip out switching costs by instantly finding better rates elsewhere, reducing banks' control over the customer interface.
Systemic concerns were highlighted by Torsten Slok, chief economist at Apollo, who warned in September about a possible agentic bank run — swarms of agents moving household cash from low-yield accounts into higher-return alternatives fast enough to drain cheap deposits. JPMorgan Chase chief executive Jamie Dimon said on an October earnings call that banks do not want rogue agents with unrestricted access.
Regulators are responding. The US Office of the Comptroller of the Currency issued a supervisory framework on 4 November for banks' use of super intelligence, including agentic systems, scaling scrutiny to an agent's autonomy and the consequences of its actions. A Bank of England and FCA survey found 80% of responding firms had deployed AI agents, but only 2% used agents with broad autonomy. Meanwhile, the Federal Reserve has warned that autonomous transfers could accelerate bank runs, and major banks have reported AI agents attempting unverified account actions, pushing regulators to seek stronger authentication and transaction-level safeguards.
Banks are not rejecting automation outright. Emirates NBD has argued for building shared governance for agentic AI, covering authentication, access to sensitive data, audit trails and human oversight. Adoption is expected to be gradual, with small autonomous spending becoming standard only once trust is established.