UK banks block third-party AI agents from customer accounts
In this article
HSBC, Barclays, Lloyds, NatWest and Nationwide have all moved to block or restrict third-party AI agents from reaching customer accounts, according to Cointelegraph, which reported the measures on 16 November. The concern is sharpest around agents able to make transactions without the customer’s involvement.
Why the lenders are closing the door on outside agents
The immediate worry is accountability. If an agent pays the wrong bill or empties an account, the dispute lands with the bank, and existing processes were built for customers making a few payments a week rather than an agent making thousands of small ones. Edgars Nemše, chief executive of the GenLayer Foundation, told Cointelegraph that even a small share of failures going wrong would generate a lot of disputes, and that the machinery for handling them still has to be built.
There is a commercial motive running alongside the risk one. Nemše argued that agents handling plain accounts would strip out switching costs, because an agent could find a better rate elsewhere as soon as it appeared online. Banks would be competing on price with a tireless shopper on the other side of the table — and losing control of the everyday customer interface in the process.
The systemic version of that worry is what Slok called an agentic bank run: cohorts of agents moving household cash from low-yield accounts into higher-return alternatives fast enough to drain the cheap deposits banks lend against. Dimon’s October remark about rogue agents landed in the same territory.
Banks want their own frameworks first
Institutions are not rejecting automation outright. Emirates NBD’s developer experience team argued in July that banks should build their own governance for agentic AI rather than let individual teams develop agents without shared controls, and its Leap framework covers authentication with core banking systems, access to sensitive data, audit trails and human oversight. Neeraj Makin, the bank’s group head of strategy analytics, expects round-the-clock banking with near-instant, atomic payments to be normal by 2036.
Ferdinand Dabitz, chief executive of AI-native banking and payments firm Augustus, expects each consumer to run dozens of agents and each business hundreds, all talking to their bank — with a human layer retained for trust, especially in commercial banking. Adoption is expected to be gradual: Nemše described an agent that reviews his finances monthly and queues payments for approval, with small autonomous spending as the next step once trust is established.
The US supervisory framework issued on 4 November is designed to help examiners judge how institutions use these systems, scaling scrutiny to an agent’s autonomy and the consequences of its actions rather than imposing new binding rules.
Why it matters
For UK customers, the practical effect is that the convenience of handing bill-paying and rate-hunting to an agent will arrive on the bank’s terms, if at all, and the liability question will shape how much autonomy anyone is granted. The same tension is playing out across every market where open banking and AI assistants were expected to disintermediate branch and app alike — a pattern worth watching given how central banks have already been repositioning their reserves, as our reporting on central bank gold buying in Q3 showed.
What to watch
How the lenders translate these blocks into formal access policies — and whether delegated access with spending limits becomes a standard feature rather than a workaround — is the next concrete marker. The OCC framework gives examiners a yardstick; the UK banks have yet to publish one of their own.
Frequently Asked Questions
Have AI agents actually accessed UK bank accounts yet?
No. Cointelegraph reported there is no evidence yet of AI agents accessing UK bank accounts, and the lender restrictions are precautionary.
What is an agentic bank run?
Apollo chief economist Torsten Slok used the term in September for a scenario where swarms of AI agents move household cash out of low-yield checking accounts en masse, draining the cheap funding banks lend against.
What did the OCC framework issued on November 4 do?
It set out a supervisory framework for how bank examiners assess institutions’ use of what the agency calls super intelligence, including agentic systems, without introducing new binding requirements.
How autonomous are deployed AI agents in finance today?
A Bank of England and FCA survey found 80% of responding firms had deployed AI agents, but only 2% used agents with broad autonomy.
Source: Cointelegraph





