Open AI Wants to Transform Banking: Why Are JPMorgan and Citi Pushing Back?

October 2026

Fintech & Payments

Shane O'Sullivan

Research Analyst

A bank’s next major competitor may never offer a current account, issue a credit card, or lend a penny. It may simply tell customers where their money would work harder elsewhere.

That possibility is beginning to concern some of America’s biggest banks. On the 16th October, The Financial Times reported that Citigroup and Bank of America had joined JPMorgan Chase in restricting third-party artificial intelligence agents from accessing their accounts. The restrictions highlight growing tensions between banks and technology companies developing AI assistants that can increasingly analyse financial information, identify better savings opportunities, and potentially initiate transactions on behalf of customers. At the centre of the dispute is whether these tools can access sensitive banking systems securely, and how much control financial institutions should retain over automated financial decisions.

AI Moves Beyond the Shopping Basket

OpenAI is already extending its reach into personal finances. Its finance services allow eligible US ChatGPT users to connect financial accounts, examine spending, and evaluate financial decisions. Critically, it cannot move money or execute transactions. Separately, OpenAI’s partnership with Visa aims to bring secure, AI-initiated payment into digital commerce, using Visa’s tokenisation, authorisation, and fraud prevention infrastructure. While these developments serve different purposes, together they demonstrate how AI assistants could evolve from providing financial information toward helping customers act on their decisions. 

Juniper Research forecasts 120 billion agentic commerce transactions annually by 2031; covering both business and consumer markets. Although this is not a forecast for automated banking transfers, it illustrates the scale of commerce that could eventually depend on trusted AI authorisation.

Are Banks Protecting Customers or Their Business Model?

The security concerns are substantial. Allowing an AI agent to review a balance is very different from authorising one to move funds. Compromised credentials, manipulated instructions, and unauthorised transactions could create complex questions over accountability.

JPMorgan chief executive Jamie Dimon has warned about rogue agents and the transfer of liability when customers give external tools account access. These restrictions do not amount to a specific ban on OpenAI; they are concerned about unapproved third-party agents more broadly. Banks also benefit from stable, low-cost deposits. An assistant that continually compares interest rates and recommends moving surplus cash could weaken the customer inertia that helps make those deposits valuable. Customers might benefit from better returns, while banks face higher funding costs or lose balances all together.

JPMorgan is not rejecting third-party financial technology outright. It supports authorised data sharing through Open Banking, while opposing AI agents logging in with a customer’s credentials. Furthermore, although Citi and Bank of America have permitted certain authorised read-only connections, they have also restricted third-party AI agents from logging into accounts or executing accounts directly. These distinctions matter for OpenAI, Meta and other AI developers; accessing financial information is only the first step towards becoming a customer’s primary financial interface.

The contrast is revealing. In January 2025, Citi predicted that consumers would use AI agents to choose financial products and execute transactions. JPMorgan, meanwhile, is piloting Smart Cash; an AI-powered service to optimise money movement within the bank. Both recognise the opportunity in automated finance, but appear more cautious when allowing independently operated third-party agents to perform similar activities.

The Battleground Is Permission

The question is not whether banks should open their systems to every autonomous tool. They should not. It is whether they can distinguish trusted, customer-authorised agents from unverified automation. Existing Open Banking arrangements offer a starting point, but moving from data access to delegated financial decisions introduces additional requirements. Banks must establish which agent is acting, what authority it holds, what it can do, and when that permission expires.

For fraud prevention providers, this presents a significant opportunity. Traditional checks assess identities, devices and transactions; agentic banking must also assess the legitimacy and limits of a machine’s authority. A familiar device or valid credential should not automatically grant unrestricted permission to transfer money.

With granular consent, spending limits, and real-time legitimate customer choice, financial institutions must balance their responsibility to safeguard funds with consumers’ ability to use trusted third-party services.  

Who Will Own the Customer Relationship?

Consumers are unlikely to surrender control of their savings overnight. Many will want to approve transfers themselves, and established banks retain considerable advantages in trust, regulation and infrastructure. Nevertheless, the competitive question is changing. If customers begin asking AI assistants where to save, borrow and invest, banks risk becoming the providers of financial products, rather than the primary source of financial guidance. 

The last decade saw banks and fintechs compete to own the customer-facing banking application. The next could see them competing to remain the institution an AI assistant recommends.

Ultimately, success will not necessarily belong to those developing the most capable AI or imposing the toughest restrictions. It will depend on which institutions can make automated finance decisions secure, transparent, and genuinely beneficial for customers. Banks may continue to hold customers' money while losing influence over where that money goes. For JPMorgan, Citi, and OpenAI, the real contest may be over who gets to act when the customer says yes.