On JPMorgan’s third quarter earnings call yesterday, CEO Jamie Dimon expressed concern about AI agents, while CFO Jeremy Barnum discussed the risks of credential sharing and screen scraping.
JPMorgan has one of the largest technology budgets in the banking sector, with plans to spend almost $20 billion on technology overall in 2026. But the C-suite executives’ comments are in no way a rejection of agentic AI as an enabling technology. The bank is already using agentic AI to automate workflows, improve employee productivity, and streamline internal processes. “Internally, we use agents to do things here, and we are going to continue to do that,” Dimon said on the call. “And we do. We will use it for customers, but our focus would always be do a great job for customers.”
But when it comes to consumer use of agentic AI, the bank said it is restricting certain methods external agents use to access customer accounts. Barnum said the bank is blocking insecure access where it can and working toward safer alternatives.
This stance has five implications banks should consider as they weigh their own attitudes toward the use of external agents.
Agentic AI could reignite the screen-scraping debate
For years, the banking industry has been moving away from screen scraping and toward more secure, API-based data sharing. However, as consumers begin using AI agents capable of navigating banking websites like human customers, the industry’s progress toward standardized, permissioned data access could face a new challenge.
Agentic AI introduces a new version of the old screen-scraping debate, as third parties may be able to access customer information without relying on bank-approved APIs. JPMorgan, which clashed with Plaid in 2025 over the terms of API-based access to consumer-permissioned financial data, has already shown interest in controlling how third parties interact with its systems. And while cybersecurity concerns are one reason the bank is wary of opening up access, the value of the data is another. Consumers’ transaction histories and account balances are valuable sources of information, particularly for AI agents capable of analyzing spending patterns, recommending financial products, and influencing financial decisions.
While some banks may seek to restrict external agents, others may view secure agent access as an opportunity to improve the customer experience. Either way, the rise of agentic AI could accelerate demand for standardized, secure, permissioned access that allows AI agents to interact with banking systems without undermining the open banking infrastructure the industry has spent years building.
Banks could lose control of the customer relationship
If consumers begin using AI assistants to manage their finances, those agents could become the primary interface between customers and their banks. For example, an AI agent might compare savings rates, and either recommend switching providers, or initiate a transfer on a consumer’s behalf without the consumer ever opening a banking app.
On the earnings call, JPMorgan’s executives acknowledged the competitive implications of AI and fintech while emphasizing the value of the bank’s established customer relationships. However, those relationships could become less influential if consumers increasingly rely on AI agents to make financial decisions. Instead of choosing a bank based on its mobile app, brand recognition, or existing relationship, consumers could delegate more of their financial management to agents that prioritize interest rates, fees, and product features. In this scenario, banks risk becoming interchangeable providers of financial products rather than the primary destination for managing money.
We’re slowly approaching a world in which consumers don’t need to take action, but can instead give their agent permission to take action on their behalf. In this new reality that doesn’t rely on consumers to take small actions like transferring funds to multiple different bank accounts per week in order to maximize yield, banks may need to compete on the quality and accessibility of their financial products, not just on their digital banking experiences.
Consumer authorization may not be enough to protect consumers
The rapid pace of AI development has raised concerns about whether safeguards can keep up. Additionally, it is a complicated enough technology that everyday consumers do not understand implications or risks, despite giving AI agents permission to access their financial accounts. Especially in the early stages of the technology, where guardrails and standards are still in formation, an agent could easily misinterpret instructions, expose sensitive information, or initiate an unintended transaction.
JPMorgan’s executives specifically raised concerns about customers sharing credentials and the potential liability associated with agent activity. When consumers share their banking credentials with an AI agent, they may be granting it access to far more information and functionality than is necessary to complete a particular task. If an agent makes a mistake, determining who is responsible for the mistake is complicated. If an AI agent transfers funds to the wrong account, for example, who is liable?
As consumer adoption increases, there will likely be growing demand for granular permissions, transaction limits, agent authentication, revocable access, and clearer liability frameworks. It is also an example of how authorization and authentication are not the same thing. An agent may be properly authenticated but still attempt an action beyond what its user intended. Know Your Agent (KYA) technologies could help bridge that gap by verifying an agent’s identity, establishing the scope of its permissions, and helping to keep its actions within authorized boundaries.
Banks may favor their own AI agents over third-party platforms
In the early 2010s, fintechs introduced a competitive tension to the banking scene. Banks were worried about losing customers to fintechs, which offered new technologies and more intuitive user interfaces. As banking-as-a-service gained popularity, however, it created new opportunities for banks and fintechs to collaborate rather than compete. Agentic AI is bringing a similar tension back as banks want to use AI to improve their services, but they may be less comfortable allowing outside agents to interact with customer accounts.
This tension could encourage banks to steer consumers back to their own websites and mobile apps by building their own financial assistants or partnering with a network of approved AI providers. Offering agentic capabilities within their existing digital banking environments will help banks offer customers the convenience of AI-powered financial management while maintaining greater control over data access, security, and the customer experience. However, this approach still has limitations, as it still may not allow consumers to use their preferred third-party AI assistants.
Third-party technology providers that can demonstrate strong security, governance, and integration capabilities may have an advantage over those relying on unrestricted website access. Instead of competing directly with banks for the customer relationship, these providers could position themselves as trusted partners that enable secure agent-to-bank interactions. Just as banking-as-a-service created new opportunities for collaboration between banks and fintechs, agentic AI could give rise to a new category of infrastructure providers focused on agent authentication, permission management, and secure financial transactions.
Trust can be both a competitive advantage and an obstacle
Banks have spent decades establishing customer trust, especially when it comes to protecting their personal information. JPMorgan’s priorities are a reflection of preserving this trust, as its executives identified trust as a competitive asset. However, as Barnum pointed out, trust doesn’t excuse banks from delivering good customer experiences.
So how can banks balance trust with improving the customer experience in the age of agentic AI? If they impose too many restrictions on third-party agents, they risk frustrating consumers who want more convenient financial management tools. However, if they allow an agent free-for-all on their sites, they risk exposing customer data, enabling unauthorized transactions, and ultimately eroding the trust they’ve worked so hard to establish.
Here’s the takeaway: the institutions that succeed may be those that balance security with interoperability, rather than simply blocking external innovation. To be clear, JPMorgan did not state that it plans to permanently shut out all external AI agents. Instead, its executives indicated that the bank is working on safer ways to enable access. “We are building tools to [protect customers from risks associated with agents] too,” added Dimon. “In the meantime, we are not gonna allow just any agent to come, outside agent to come inside the company because it is just too risky for clients and for JPMorgan.”
