Holy Toledo! If there was ever a no-brainer statement, this is it!
Don’t give AI agents your bank account number! Those busy little beavers could do – who knows what – with your stash of cash.
I think this has Macroeconomic significance because all the AI companies need to make money to cover their hundreds of billions of dollars of investment. And they will do their darndest to convince the public that AI agents can do all kinds of nifty tasks for them, such as answering their e-mails and paying bills from their bank accounts.
Please Don’t Let an AI Bot Log Into Your Bank Account
Banks are warning against giving too much access to AI agents.
by Dashia Milden, cnet.com, October 23, 2026
Whatever you do with AI, don’t give AI bots your banking information.
Agents like Meta’s Muse, OpenClaw and OpenAI’s dots are all the rage in the AI world these days. They promise to go beyond a chatbot that gives you information, becoming a tool where you can tell it to do something for you and it goes and does it.
When you first sign up for an AI agent, you may be asked to grant permission for it to access your apps, browsers and data so it can work with them. That might put your money at risk in unexpected ways.
“The intention might be something simple, like managing a calendar. But that access can give the agent the ability to do much more than the person intended,” said John Strand, the owner of the cybersecurity firm Black Hills Information Security. While you may be giving the agent access to simple smart home routines or getting help planning a trip, your banking information may also be accessible…
One of those challenges is that they don’t always behave in the way you expect or do what you want them to…
There are also plenty of reasons to be worried about the privacy of data shared with agents. Your agent might not meddle in unwanted ways with your money, but it could compromise your login info, letting someone nefarious meddle with your money… [end quote]
Now, I would rather have cranked windows in my car than electric windows, but this takes risk to a whole new level. I want my AI to be a super-sophisticated search engine or spreadsheet, not to be able to act on my behalf.
I asked Gemini’s opinion and it pointed out that malicious code could hijack the AI agent. “Unlike a traditional script, AI agents read unstructured text and data. If an agent accesses your email or browser to pay a bill and encounters a malicious string of text hidden on a web page or invoice (a prompt injection attack), the AI can be tricked into overriding its original instructions—such as rerouting money or exfiltrating credentials.”
Bottom line: No way, Jose.
Wendy