Bank of America Tells Customers: Don't Give Your Password to Your AI Assistant

Imagine handing your house keys to a stranger because they promised to water your plants while you're away. That's roughly the bargain millions of people are now making with their AI assistants, except the house is their bank account.

Bank of America wants customers to stop doing that. In an email blast titled "Protecting your finances in the AI World," the bank made a simple plea: never share your username and password with another person or outside service, "including an AI-enabled tool or agent that offers to act on your behalf." A spokesperson for the bank confirmed the email's authenticity after it surfaced on social media, posted by Bryan Toporek.

Read between the lines and you can almost hear the exasperation. Banks have spent decades training customers to guard their credentials like house keys. Now a new category of software, personal AI agents, is asking people to do the opposite: hand over the keys so the assistant can pay bills, compare prices, or move money without you lifting a finger.

The appeal is obvious. Nobody enjoys manually comparing flight prices at 2 a.m. or triple checking a utility bill. But convenience has a cost, and in this case the cost is control. Share your login with an AI tool and you lose the ability to track exactly what it's doing with your money in real time. Worse, if that tool has a security flaw, your credentials could leak out to the wrong people entirely.

The Mechanics of the Risk

Think of your banking password as a master key to a vault. Normally, only you hold it. When you give that key to a third party, even a well-intentioned one, you're trusting that they'll use it responsibly and that nobody else can steal it from them. AI assistants are software, and software has bugs. A security flaw in an AI agent could let hackers intercept your credentials as they pass through, giving financially motivated criminals a direct line into your account.

There's a second, quieter risk too. Banks run fraud detection systems that watch for unusual login patterns and behavior. If an AI agent starts logging in and transacting on your behalf, the bank's systems might flag that activity as suspicious, the digital equivalent of a security guard assuming you're a burglar because you used your emergency key at an odd hour. That can result in a frozen account, which is deeply inconvenient regardless of whether anything fraudulent actually happened.

This warning didn't appear in a vacuum. Banks are reportedly already grappling with unauthorized AI agents accessing customer money, according to recent reporting from Axios. At TechCrunch Disrupt, panelists discussed how banks, payment providers, and processors are still trying to figure out how to safely let AI tools into the financial system at all, particularly for agentic payments, where an agent autonomously buys something at the best price or the ideal moment.

That's the tension underneath all of this. The financial industry sees real value in agentic commerce. An AI that monitors prices and buys when they dip, or renegotiates a subscription on your behalf, could genuinely save people money and time. But the infrastructure to do that safely, with proper authorization protocols instead of simple password sharing, isn't fully built yet. Customers are improvising in the meantime, often by just typing their login into whatever assistant asks for it.

And here's the part that should give everyone pause: if something goes wrong, you're the one holding the bag. Bank of America's email states it plainly. "If you do share your credentials, keep in mind that you're responsible for activity resulting from that access." This isn't unique to Bank of America. Account agreements at most major banks, including U.S. Bank, hold customers responsible for anyone they voluntarily give their login to, regardless of whether that recipient is a person or a piece of software.

That liability structure makes sense in a world where "sharing your password" meant giving it to your spouse or a trusted employee. It gets murkier when the recipient is an AI system whose inner workings, data handling practices, and security posture you likely don't fully understand. You're agreeing to accept responsibility for a black box's behavior, and most people signing up for a shiny new AI assistant probably aren't reading it that way.

Why It Matters

The gap between how fast AI agents are being adopted and how slowly financial institutions can build secure rails for them is where the real danger lives. People are drawn to the promise of AI handling tedious financial tasks, and that demand isn't going away. Meanwhile banks are still working out basic questions: how do you verify an AI agent's identity, how do you limit what it can access, how do you reverse a mistake it makes.

Until those answers exist, the responsible path is the unglamorous one. Treat your banking password the way you'd treat a key to a safe deposit box: something you don't hand to anyone, software included, no matter how helpful it promises to be. The convenience AI assistants offer is real, but so is the exposure. For now, the safest agentic payment is still the one you authorize yourself, one transaction at a time, with your own hands on the keyboard.