UK Banks Restrict AI Agents Over Security and Competition Concerns
Several UK banks, including HSBC, Barclays, Lloyds, NatWest and Nationwide, have blocked or restricted third-party AI agents from accessing customer accounts, particularly when the systems could make transactions without a customer’s involvement. According to Cointelegraph, lenders are concerned about the consequences if an automated agent mishandles money and who would be liable, although there is no evidence that AI agents have accessed UK bank accounts so far. Banks may also be concerned that agents could make it easier for customers to switch to better offers. Edgars Nemše, chief executive of GenLayer Foundation, said agents could take over routine tasks such as paying bills, transferring money and comparing rates. If agents can quickly identify more attractive offers online, he said, banks could face greater competition and lose customers. In September, Apollo chief economist Torsten Slok warned of a possible “agentic bank run,” in which agents move household cash from low-yield accounts to higher-return alternatives. That could reduce banks’ access to low-cost funding used for lending, with the speed and scale potentially affecting the broader financial system. JPMorgan Chase CEO Jamie Dimon also cautioned against unrestricted access, saying on an October earnings call: “You don’t want rogue agents.” The adoption of AI agents in banking is expected to develop gradually, but some industry leaders anticipate wider use in the coming years. Emirates NBD strategy analytics head Neeraj Makin predicts around-the-clock banking and near-instant payments will become normal by 2036, with agents handling more financial decisions. Augustus CEO Ferdinand Dabitz said consumers and businesses could each use multiple agents to communicate with banks, while human oversight would remain important, especially in commercial banking. In July, Emirates NBD’s developer experience team said financial institutions should establish shared frameworks to govern authentication, data access and agent actions, supported by monitoring, audit trails and human oversight. Reliability remains a concern: a recent Bank of England and FCA survey found 80% of responding firms had deployed AI agents, but only 2% used agents with broad autonomy. Nemše said early uses could include reviewing finances, scheduling payments and eventually authorizing small transactions, while larger amounts would require approval. He also called for banks to provide delegated access, permissions and spending limits so customers do not have to share login details. Existing dispute systems may struggle to handle large volumes of agent-initiated transactions. The U.S. Office of the Comptroller of the Currency issued a supervisory framework on Nov. 4 to help examiners assess banks’ use of “super intelligence,” including agentic systems. The framework does not create new binding requirements; scrutiny is to reflect an agent’s potential impact, autonomy and level of human oversight.
