It sounds strange: AI assistants logging into people’s bank accounts. You might assume hackers are behind it. But in these cases, users gave their own AI assistants their banking usernames and passwords, and the assistants carried out transactions, sometimes without checking with the account holder first.

An AI agent is simply a software program that carries out tasks for you. It’s an autonomous agent that can take action and make decisions by itself, without specific instructions. When the task involves banking, that can be very risky. And when actual money is involved, everyone pays attention much more closely.

As AI agents become more sophisticated, organizations need professionals who can use these technologies responsibly, manage their risks, and determine where human oversight is needed.

How banks respond to this will be telling. One bank has been blocking these assistants outright. Another has pointed customers toward open banking, a safer way of sharing access: instead of handing over your password, you give permission through the bank itself, and the other side gets limited, revocable access. One approach shuts the door, while the other builds a better door.

Most people expect AI agents to get access to banking information sooner or later, and widely. That shifts the focus to a different set of questions: How broadly is an agent’s power to act defined? Can that power be limited? How should controls on an agent be set and when should they be tightened or loosened? These are the kinds of questions we are interested in exploring.

Good Guardrails Are Harnesses, Not Fences

It may seem easier to simply ban agents completely. But people will always prefer the most convenient way of doing something, so a complete ban may not achieve the bank’s objectives. A partial ban may work better—read-only access, for example, where an agent can look but cannot move money on a customer’s behalf. Access granted through the bank itself is revocable, so it can be taken back in a moment.

In our research, we used a market simulator to let several types of AI agents trade. We found that defensive agents that put friction in attackers’ way can be very effective. In one simulation, a defensive agent standing ready with numerous standing orders reduced the price swings caused by attacking agents by about 90%. Adding friction and limits on agents, therefore, may achieve a bank’s objectives as well. Often, the mere presence of a limit changes behavior.

Friction can also be smaller and more strategically placed. In the simulation, a small cost on each trade reduced the total number of trades and kept harmful trading in check. Similarly, your banking app could let you set limits on transfers, so that your financial assistant cannot make large transfers without your approval.

Focus on Transactions, Not Individuals

In most cases, banks do not need to know the name of each assistant a customer uses. More likely, banks spot gaming by looking for inconsistent and irregular transactions. In our simulation, we were able to detect gaming by looking at aggregate transaction data. In most cases, transaction data reflects an individual’s normal and expected behavior. Looking at transaction data in aggregate is both effective and efficient. If the data raises concerns, the first step should be to ask questions, not to assume the individual or agent is behaving maliciously.

What You Can Do Now

There are habits that protect your privacy and security. One of the most important is never to share your banking or other financial account passwords with any app or AI assistant, no matter how helpful or convenient it may seem. If you want to give an app access to your bank accounts, use the secure access your bank provides. Regularly review the access you have given to apps and services, and remove what you no longer use. Contact your bank to find out if and how it regulates the use of AI assistants. The more customers ask these questions, the sooner banks will develop policies to address customers’ privacy and security concerns.

Why This Matters

Understanding these questions and regulations will become essential for managers and senior leaders. They will shape how companies use artificial intelligence (AI) and other automation technology. Georgetown University’s School of Continuing Studies recognizes the important role of automation and AI in the workforce and has developed courses to prepare students to work with these technologies.

For professionals looking to build these capabilities, the Master’s in Strategic Leadership & AI prepares students to use AI as a strategic leadership tool, make informed decisions, and guide organizational change. The program emphasizes responsible AI adoption while keeping the human elements of effective leadership at the center.

AI assistants are not going away. They will get more capable, and more of us will use them for everyday tasks, including financial ones. Creating AI systems to perform financial transactions raises many legal and ethical issues—for example, should AI systems only perform transactions at a human’s instruction? It is not a job for banks alone. It is something all of us, as customers and as professionals, have a stake in getting right.