Is It Safe to Give an AI Agent Your Bank Login or Card? Banks Are Already Blocking Them. What to Share and What Never To

Is It Safe to Give an AI Agent Your Bank Login or Card? Banks Are Already Blocking Them. What to Share and What Never To

By Sergei Ponomarev β€’ 2026-11-16

The short answer: never give an AI agent your bank username and password, be careful about letting one use your main card, and do not paste full account numbers into a chatbot. Use the safer routes instead: read-only connections your bank approves, virtual cards with limits, and documents with the numbers blacked out. The longer answer is that the question is no longer hypothetical, and the banks have started to push back.

In late October, CBS News, following a report by Axios, described what it called an industry-wide problem: major US banks detecting activity from third-party AI agents inside customers' online accounts. In at least one case, an agent called a bank with a complaint without first asking the account holder. The agents were not hacking anything. Customers had simply given them their login details.

At the same time, people are using these assistants more than ever. In a SEON survey of 2,026 US adults, 68% said they use an AI agent at least once a week for a personal task. The question of what to let them touch has arrived in millions of households at once.

What the banks are saying

WhoWhat they said, as reported
Chase (spokesperson)"Customers should not give their username and passwords to anyone." Chase says it supports secure, customer-authorized data sharing through open banking providers
JPMorgan CEO Jamie Dimonthe bank is not going to allow just any outside agents inside, "because it's just too risky for clients and for JPMorgan"
Charles Schwabits CEO said the firm has been blocking unauthorized agents, according to Reuters
Morgan Stanleysaid it had not seen unauthorized agent activity on its platforms
White House AI adviser David Sacksdrew a line between agents a user authorizes and "rogue" ones, and suggested banks offer read-only access

The dispute is less about whether agents are useful and more about how they get in. Handing an agent your password gives it the same access you have: it can see everything and, in principle, do everything. That is exactly what banks' terms of service usually forbid, and it is why several banks are treating these agents as unauthorized.

What ordinary people actually think

Three surveys published this autumn draw the same line: people are happy for AI to protect their money, much less for it to spend it.

FindingSurvey
77% are concerned about giving an AI tool permission to move money or make payments for themAmerican Bankers Association, as reported by NewsNation
57% are comfortable with an AI flagging or freezing a suspicious charge without asking firstFIS Consumer Pulse, with Kantar
50% want to approve every transaction an AI assistant makesFIS
50% are uncomfortable with an AI agent using their card details to complete a purchase they asked forSEON, 2,026 US adults
Only 15% would share financial information with an agentSEON
43% are not confident they could tell whether an agent's request for information or access is legitimateSEON

When asked what would make them trust an AI with money, the answers were practical. In FIS's survey, 58% chose a refund guarantee for mistakes and 56% chose access to a human who can review and override decisions. In SEON's, half wanted human support if something went wrong and 42% wanted the ability to reverse mistakes.

And on who should pay when an agent's transaction turns out to be fraud, 65% of SEON respondents with an opinion said the company behind the agent, against 14% who said themselves.

The real risks, in plain words

1. You may lose the protection you assume you have. Most people rely on their bank to refund fraud. If you broke the bank's terms by sharing your login, a dispute can become harder. Read your bank's terms before connecting anything with your password.

2. An agent can make a mistake with real money. It can misread an instruction, buy the wrong item, pay the wrong bill or, as in the case CBS described, contact your bank on your behalf without asking. Unlike a typo in an email, a payment is not always reversible.

3. A scammer can pretend to be your agent. If you get used to an AI asking for confirmations, codes or access, it becomes easier for a fake one to do the same. In FIS's survey, 35% had been targeted by an AI impersonation scam or knew someone who had, and 22% had been targeted personally; most of those received fake-voice phone calls.

4. Whatever you paste may be stored. A bank statement uploaded to a chatbot can include your full name, address, account number and every merchant you use. Depending on the service and your settings, it may be kept and used to train future models.

What to share with AI, and what never to

Safe to shareShare with careNever share
A budget with amounts and categories but no account detailsA bank statement with account numbers, address and card digits blacked outYour bank username and password
Questions about a bill, a fee or an investment termYour income and debts, for planning, on a service you trustOne-time codes and security answers
A screenshot of a price or a policyRead-only access through a connection your bank approvesYour full card number, expiry and security code typed into a chat
Drafts of letters to your bank or insurerA virtual card with a low limit for a specific purchaseRemote control of your banking app or computer while you are logged in

One rule covers most of it: share what you would show a helpful stranger, never what would let them act as you.

Safer ways to let AI help with money

Use read-only connections your bank approves. Many banks let you connect budgeting and planning apps through approved data-sharing services, with permissions you can see and revoke. That gives an app your transactions without your password, and lets it look without moving money. Chase pointed to this route specifically.

Use virtual or limited cards for purchases. Many card issuers and payment apps offer single-use or merchant-locked virtual card numbers with a spending cap. If an agent must buy something, give it one of those, not your main card. Payment companies are building this in: FIS says its agent tools let an AI pay with its own tokenized credential governed by limits set by the consumer and the card issuer, rather than the consumer's primary card.

Keep a human approval step. Set any agent you use to ask before it pays, books or sends money. Half of Americans already want exactly that, and it costs you a tap.

Turn on alerts. Transaction alerts on your phone are the cheapest protection there is. If an agent or a scammer does something unexpected, you see it in seconds.

Black out the numbers before you upload. If you want help reading a statement, remove your account number, card digits and address first. The AI does not need them to explain a fee.

In the EU, national rules on automated assessments for consumer credit are due to apply from 20 November 2026; the overview is in AI regulation in banking.

What this means for you

If you use ChatGPT, Claude, Gemini or another assistant for money questions, keep doing it: understanding a bill, building a budget or preparing questions for an adviser are some of the most useful things these tools do. Just keep your credentials out of it. If you are still deciding which assistant to use at all, how to start using AI covers the basics.

If you are tempted by an agent that promises to manage your money, ask three questions first. How does it connect to my bank, password or an approved connection? Can it move money, or only read? And who pays if it makes a mistake? If the answers are vague, so is your protection.

If you are older, or helping a parent, be especially careful with anything that asks for codes or access. In FIS's survey, only 3% of people aged 70 and over were ready to hand day-to-day money management to an AI, against 61% of millennials, and scammers know which group is less used to these tools.

The same pattern runs through other parts of everyday finance. AI is already changing what you pay for things, as I set out in why some McDonald's are more expensive, and how stores price you personally in surveillance pricing. On the business side, how agents will buy things for you and who gets paid is in agentic commerce.

The honest take

What strikes me about this story is how quickly convenience ran ahead of the rules. People handed their bank passwords to AI agents because it was the easiest way to get help, and banks discovered it in their logs. Nobody designed it that way; it just happened, one helpful shortcut at a time.

The good news is that the safer version already exists in pieces: read-only connections, virtual cards, approval steps and alerts. And the public instinct is sound. Let AI watch your money before you let it spend it, keep a human in the loop, and never give anything that lets software become you.

So, before you connect the next app: does it need your password, or just your permission?

Frequently asked questions

Is it safe to give an AI agent my bank login?

No. Banks such as Chase tell customers never to share their username and password with anyone, and sharing them may break your bank's terms. Use read-only connections your bank approves instead.

Is it safe to give AI my bank statement?

Only after removing your account number, card digits and address. The AI can explain fees and spending without them, and whatever you upload may be stored by the service.

Can an AI agent use my credit card?

Technically yes, but half of Americans are uncomfortable with it. If you do, use a virtual card with a low limit, require approval before each purchase and turn on transaction alerts.

Are banks blocking AI agents?

Some are. CBS News reported that major US banks are detecting unauthorized AI agent activity in customer accounts. JPMorgan's CEO said it will not allow just any outside agents in, and Schwab's CEO said it has been blocking unauthorized agents.

Who pays if an AI agent makes a fraudulent or wrong payment?

There is no single rule yet, and it depends on your bank, card issuer and how the agent was connected. In SEON's survey, most people with an opinion said the company behind the agent should pay.

What should I never share with AI?

Passwords, one-time codes, security answers, full card numbers with expiry and security code, and remote access to your banking app or computer while you are logged in.

Sources: CBS News: AI agents are roaming around consumers' bank accounts; FIS: U.S. adults would let AI protect their money before letting AI spend it; CFOtech: SEON survey, US consumers embrace AI agents but fear money risk; NewsNation: three-quarters of Americans won't let AI access their wallets.

Share this article: