Bank of America has told customers not to share their banking username and password with any AI tool, even one that offers to act for you. That Bank of America AI agent warning reached customers by email in late October, and it matters because millions of people now let AI assistants shop and book things for them.
The short version: keep your login to yourself, and give an agent only the narrow access it needs.
Here are the key facts before the detail.
- The warning went out around October 27, 2026, according to TechRadar, which cites TechCrunch.
- It covers any AI-enabled tool or agent that offers to act on your behalf.
- Other big US banks are tightening access too, per Axios.
- Your best protection today is approval prompts and virtual card numbers.
Those four points are enough if you’re in a hurry. The rest explains why banks are worried and what to do instead.
What the Bank of America AI agent warning actually says
The bank’s message is simple. Don’t give your username and password to anyone, including an AI tool that promises to handle tasks for you. TechRadar reports the bank framed this as a safety rule, not a ban on using AI.
The email landed after a busy month. A week earlier, Axios reported that unauthorized AI agents were showing up inside customer accounts at several big banks. In most cases, customers had handed over their logins, which breaks most banks’ terms of service.
Why banks are nervous about AI agents
The core problem is that banks often can’t tell who did what. Per the Axios report, as relayed by OfficeChai, a bank may not know whether you gave the order, an approved tool did, or an agent went beyond what you allowed.
That makes fraud harder to spot and refunds harder to settle. If a purchase happened only because you gave an agent your credentials, TechRadar notes the customer may carry most of the responsibility.
There’s also a lockout risk. Agent activity can look like unauthorized access, so a bank may freeze your account to be safe.
Trust is the other gap. A Visa-commissioned Harris Poll found 72% of consumers have used an AI assistant, but only 23% would let one pay for something on its own, per Techlicious. Many shoppers already feel what the banks are saying out loud.
Some stores have pushed back as well. Techlicious reports Amazon blocked Meta’s Muse agent in September, saying it didn’t identify itself as AI and stored login details.
Which banks are acting
Axios named Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and Citi. Their responses differ, so here’s a quick comparison of what was reported.
| Bank | Reported response |
|---|---|
| Morgan Stanley | Has started blocking third-party agents on its platforms. |
| JPMorgan, Bank of America, Citi | Tightening access for outside agents. |
| Most banks overall | Moving toward read-only agent access, where an agent can view data but not act. |
| Chase (spokesperson) | Told customers not to hand their username and password to anyone, “AI agent or otherwise.” |
Read-only access is the likely middle ground. It lets helpful tools see your spending without letting them move money.
How to use AI helpers without risking your money
You don’t have to avoid AI assistants. You just need to set the limits yourself. This table shows which setups carry more risk.
| Action | Safer? | Why |
|---|---|---|
| Give an AI agent your bank username and password | No | The bank can’t tell you from the agent, and you carry most of the risk. |
| Let an agent view balances through a bank-approved, read-only link | Mostly | It can see data but can’t move money. |
| Let an agent shop with a virtual card number | Yes | You can shut the number off instantly without touching your real account. |
| Let an agent buy without asking you first | No | A hijacked or confused agent could spend before you notice. |
In September, Bank of America joined Capital One, NatWest, ING, ASB and Commonwealth Bank of Australia on shared safety principles for AI agents, Techlicious reported. Their advice matches the table: require approval before any purchase, and ask your bank about virtual card numbers.
If you already shared a login with an AI tool, change that password now and turn on two-step sign-in. Then check your recent transactions for anything you don’t recognise.
Our look at Amazon’s AI shopping agents shows why these tools still make mistakes. And if you connect AI to personal accounts, our ChatGPT iCloud plugin guide covers how to do it more safely.
Frequently Asked Questions
Is it safe to give ChatGPT or another AI my bank password?
No. Bank of America and Chase both say not to share your login with any AI tool. Use approval prompts and virtual cards instead.
Will my bank refund me if an AI agent makes a bad purchase?
Don’t count on it. If you gave the agent your credentials, reports suggest you may carry most of the responsibility. Check your bank’s own policy.
What is a virtual card number?
It’s a stand-in card number linked to your real account. You can switch it off instantly if something looks wrong. Techlicious says Capital One and Citi offer them in their apps.
Can my bank lock my account because of an AI agent?
It’s possible. Agent activity can look like unauthorized access, and banks may freeze accounts to stay safe.
Our take
Treat an AI agent like a helpful stranger with fast hands. Let it research, compare and draft, but keep the final payment step in your own hands, and never hand over your bank login. Until banks and AI companies agree on proper bank-approved access, that one habit protects you better than any setting.
Leave a Reply