As major banks move to restrict AI assistants from accessing customer accounts, they face a difficult question: How do you protect customers from untrusted agents without preventing them from using the next generation of digital banking?
For decades, banks have invested billions of dollars making it easier for customers to manage their money digitally. They built websites, mobile applications, APIs, and sophisticated authentication systems. They encouraged customers to move from branches to digital channels, promising greater convenience, speed, and control.
Now, a new generation of technology promises to make banking even easier. And some of the world’s largest banks are trying to keep it out.
On October 16, the Financial Times reported that Citigroup had joined other major US financial institutions, including Bank of America and JPMorgan Chase, in restricting third-party AI agents from accessing customer accounts.
The stated concern is security. AI agents could access sensitive information, initiate unauthorized transactions, or move money without appropriate customer consent.
These are legitimate risks. Banks have an obligation to protect their customers and their assets.
But the Financial Times highlighted another concern: AI assistants could become remarkably effective at helping customers make better financial decisions, including moving billions of dollars from low-interest deposits to higher-yielding alternatives.
And that raises a much bigger question.
Are banks blocking AI agents because they threaten customer security, or because they threaten the economics of traditional banking?
The answer may be both. But regardless of the motivation, blocking AI agents is unlikely to be a sustainable long-term strategy.
What Happens When Your AI Assistant Becomes Your Financial Advisor?
Imagine you have $100,000 sitting in a bank account earning 0.1% interest.
Your personal AI assistant reviews your finances and identifies another federally insured account offering 4%.
It calculates that you’re potentially losing $3,900 annually by leaving your money where it is.
Then it asks a simple question:
“Would you like me to move your money to an account where it can earn more interest?”
You say yes.
For the customer, this is a fantastic service. No research, no comparison websites, no complicated forms. Just an intelligent assistant looking after their financial interests.
For the bank holding those deposits, it’s a very different story.
Traditional banking relies partly on customers keeping money in accounts that pay relatively little interest. Many customers simply don’t have the time, knowledge, or motivation to continuously optimize their finances.
AI agents could change that equation.
An assistant doesn’t get tired. It doesn’t forget to compare interest rates. It doesn’t find switching banks inconvenient. And unlike a bank’s own financial advisor, it may have no commercial incentive to recommend that the customer keep their money where it is.
The implications extend far beyond deposits.
AI assistants could compare credit card fees, negotiate loan terms, identify unnecessary insurance coverage, challenge charges, and recommend competing financial products.
They could fundamentally change the relationship between banks and their customers.
For the first time, every consumer could potentially have a financial advisor working continuously in their own interests.
That should be an exciting opportunity for the banking industry.
Instead, some banks appear to be approaching it primarily as a threat.
Blocking AI Agents Is Easier Said Than Done
There is another problem with a strategy built around blocking AI agents: reliably identifying them is difficult.
Traditional bot detection was designed largely to distinguish human users from automated software. It looks for signals such as unusual navigation patterns, automated browser behavior, abnormal request rates, and the absence of typical human interactions.
These techniques remain valuable, particularly against conventional automation and known malicious bots.
But modern AI assistants present a different challenge.
Some operate through browsers, interacting with websites much like human users. Others use computer-use capabilities to navigate applications, click buttons, fill out forms, and complete workflows.
An agent may operate within a customer’s existing authenticated session, using the same browser, device, and network connection.
In those circumstances, identifying the underlying agent can be substantially more difficult than identifying traditional automation.
And as agents become more capable, the distinction between human and agent behavior may become increasingly difficult to establish from behavioral signals alone.
This doesn’t mean banks cannot detect or block AI agents. They can, and some techniques will continue to be effective.
But it does mean that blocking agents is an ongoing technical contest, not a permanent solution.
Even more importantly, successfully detecting an AI agent doesn’t answer the most important questions.
Is this agent acting with the customer’s permission? Is it attempting a legitimate financial task? Is it accessing information or initiating a transaction? Has the customer authorized that specific action?
A blanket blocking policy treats these very different situations as though they present the same risk.
They don’t.
The Customer Is Changing. Banks Need to Change With Them.
Think about what happened when mobile banking first emerged.
Customers wanted to access their accounts from smartphones. Banks had to adapt their security models to a new environment, introducing mobile authentication, device intelligence, transaction monitoring, and new fraud controls.
The solution wasn’t to permanently prevent customers from using mobile devices.
It was to make mobile banking safe.
AI assistants represent another fundamental change in how customers interact with digital services.
Today, customers typically visit a bank’s website or mobile application and perform tasks themselves.
Tomorrow, many of those interactions may be initiated and executed by AI assistants acting on their behalf.
The customer remains the customer. What changes is the interface through which they interact with the bank.
This creates a strategic dilemma.
If customers increasingly prefer delegating routine financial tasks to AI assistants, what happens to banks that refuse to support that interaction model?
Will customers abandon their assistants to accommodate their banks?
Or will they eventually favor financial institutions that make it easier to use the technology they prefer?
The answer isn’t certain, and adoption will take time. Customers may be reluctant to grant AI assistants control over their finances, particularly when large transactions are involved.
But banks should be careful about betting their long-term digital strategy on customers continuing to interact with financial services exactly as they do today.
The next generation of digital banking may not begin with a customer opening a banking app. It may begin with a customer telling an AI assistant what they want to accomplish.
Banks need to be prepared for that possibility.
The Answer Isn’t Unrestricted Access. It’s Intelligent Control.
None of this suggests banks should simply open their applications to every AI agent.
Quite the opposite.
AI agents introduce risks that existing authentication and fraud controls were not necessarily designed to address.
An agent might have legitimate access to a customer’s account but attempt actions the customer never intended. It could be manipulated through malicious instructions, make incorrect decisions, or expose sensitive information.
A compromised agent could be especially dangerous because it may already possess valid credentials or access to an authenticated session.
And an agent that successfully authenticates as a customer isn’t necessarily authorized to perform every action that customer could perform.
Authentication establishes access. It does not establish unlimited authority.
From Blocking Agents to Building Trust
For banks to safely embrace AI-driven customer interactions, several capabilities will become increasingly important.
First, banks need visibility into agentic activity across their digital channels. They need to understand when an AI agent may be interacting with their applications, which platform it may be associated with, and how it behaves.
Second, they need mechanisms to establish authorization. A legitimate agent should be able to demonstrate what the customer has permitted it to do. That requires more than recognizing an agent’s technical fingerprint. It requires trustworthy identity, delegation, and consent mechanisms.
Third, banks need continuous risk assessment. Trust cannot be established once at login and assumed to remain valid throughout an entire session.
An agent that begins by reviewing an account balance may later attempt to modify account details, register a new payment destination, or initiate a transfer. Each action introduces different risks.
Finally, banks need the ability to respond dynamically.
A trusted assistant reading transaction history might proceed without interruption. A high-value transfer might require explicit customer confirmation. Suspicious activity might trigger additional verification, restricted permissions, or immediate termination.
These decisions should consider the agent, the customer, the requested action, the surrounding behavior, and the potential consequences.
This is the direction we believe the industry needs to move toward at Transmit Security.
Our work in Agentic Trust is focused on helping organizations understand the AI agents interacting with their customer-facing applications, assess their behavior and risk, and provide intelligence that supports appropriate security and business decisions.
But this is bigger than any single technology.
It requires banks to rethink how they establish trust when the entity interacting with their application is no longer necessarily a human.
There’s Also a Business Opportunity Banks Shouldn’t Ignore
The discussion around AI agents in banking is currently dominated by risk.
Fraud. Unauthorized transactions. Credential exposure. Deposit flight.
These are important concerns, but they represent only one side of the equation.
AI assistants could also become a powerful new customer acquisition and engagement channel.
Imagine a customer asking an assistant to find the best mortgage, credit card, or savings account for their circumstances.
Banks that make their products accessible, understandable, and easy for legitimate agents to evaluate could become more competitive in this new environment.
Banks that make those interactions unnecessarily difficult could become less visible to customers who increasingly rely on assistants to make decisions.
There is also an opportunity to serve existing customers better.
A bank could allow a trusted assistant to help customers manage their finances, discover relevant products, complete routine servicing tasks, and navigate complex financial decisions—all within clearly defined security boundaries.
Instead of treating AI assistants exclusively as a threat to the customer relationship, banks could use them to strengthen that relationship.
The critical requirement is control.
Banks must be able to distinguish legitimate, customer-authorized activity from fraud, manipulation, and abuse.
The opportunity isn’t to eliminate security controls. It’s to make those controls intelligent enough to support an entirely new way of doing business.
You Can’t Build the Future of Banking by Blocking It
Banks have good reasons to be cautious about AI agents.
The technology is evolving rapidly. Security risks are real. Standards for identity, delegation, consent, and liability are still developing.
Temporary restrictions on high-risk activities may be entirely appropriate while those issues are addressed.
But a temporary security measure should not become a permanent digital strategy.
History suggests that when technology gives customers meaningful improvements in convenience, choice, and control, businesses eventually have to adapt.
AI assistants could become one of the most significant changes in the relationship between consumers and financial institutions since the introduction of online and mobile banking.
Banks can respond by trying to preserve the existing interaction model.
Or they can invest in the intelligence, identity, authorization, and control capabilities needed to support the next one.
The choice will affect much more than cybersecurity. It could influence customer loyalty, competition, product distribution, and the economics of banking itself.
The banks that succeed in the AI era won’t necessarily be the ones that build the highest walls against AI agents.
They’ll be the ones that learn how to recognize them, understand them, control what they can do—and turn them into a trusted new channel for serving customers.
