Banks face a new business model
The bigger challenge is the emergence of agentic AI, which can plan and execute tasks with limited human supervision. Customers could eventually authorise AI agents to purchase goods, compare savings rates and move money between accounts without opening individual apps.
This puts banks in a difficult spot. Their payment systems and fraud checks are designed around how people normally transact, including confirming that a customer has approved a payment. But an AI agent handling a small business’s finances could make thousands of payments across merchants and currencies, setting off fraud alerts simply because its activity looks unusual.
The financial stakes are also significant. Bank of America has reported about $800 million in annual savings and revenue benefits from generative AI after an initial investment of $400 million. Société Générale sees potential AI savings of up to €600 million, while Commerzbank expects AI to deliver €500 million in annual benefits by 2030.
Yet banks may not retain all these benefits. AI agents could help customers move deposits into higher-interest accounts, putting pressure on lenders that rely on relatively cheap deposits to support their earnings.
Cybersecurity risks are rising
The transition also raises concerns about who controls autonomous financial activity. On 13 October, regulators in the United States and Europe stepped up scrutiny of AI risks in banking, including potential cyberattacks and threats to financial stability.
US authorities, including the Federal Reserve, have questioned banks about their use of advanced AI models, whether they deploy technology developed in China, and how they control and audit autonomous systems. In Europe, the Bank of England and European Central Bank are pressing for stronger cyber-recovery plans and safeguards against disruptions.
The challenge is that AI agents could act faster than banks can detect or contain harmful activity. Independent backup systems and stronger recovery arrangements may help, but implementing them can be costly.
Why banks cannot afford to wait
Banks face a balancing act: they must prevent malicious agents from accessing accounts while allowing authorised agents to transact on customers’ behalf. Britain's AI minister, Kanishka Narayan, warned this week that cyberattacks on the UK by rogue AI agents were likely, if not inevitable.
The response will require banks, regulators and AI companies such as OpenAI and Anthropic to work together on safeguards, permissions and accountability.
The central question is no longer simply how banks can use AI to improve existing services. It is whether they can remain central to customers’ financial lives when AI agents increasingly choose products, move money and make payments for them.