Hand your AI assistant the keys to your bank account and Citi has a message for you: whatever happens next is your problem.
Citi, the third-largest commercial bank in the United States, updated its Online User Agreement on November 18 with a new section on third-party AI agents. The terms say that if you let an AI agent into your account, everything it does counts as authorized by you. And if that agent leaks your credentials, or gets hacked, the bank will treat whatever a thief does with them as authorized too.
It’s the sharpest warning yet from a major bank about a habit that’s spreading fast. AI agents can now log in to your accounts, pay bills, and move money. The banks, it turns out, want no part of the risk.
What Citi’s new terms actually say
The update adds a section titled “Use of Third-Party Artificial Intelligence” to Citi’s Online User Agreement. It covers any third-party AI agent used to access accounts or use digital services through Citi Online or Citi Mobile.
Three parts of the new language matter most.
First, access is on you. If you give an agent your credentials or otherwise permit it into your account, Citi treats the agent’s actions as authorized by you, even when those actions conflict with your own instructions.
Second, leaks are on you too. If the credentials you gave the agent get shared, leaked, or passed along to other people or AI agents, Citi says it will treat any later access and actions taken with them as authorized by you, except where applicable law says otherwise.
Third, Citi keeps the kill switch. The same section lets the bank limit, block, suspend, or end an agent’s access at any time, without notice.
Citi’s own AI tools are exempt. The bank’s terms note that AI agents it provides can be used without the extra liability.
A Citi spokesperson told TechCrunch the update was meant to spell out the danger plainly. “As third-party AI tools become more widely available, we want customers to understand the risks associated with sharing banking credentials or authorizing those tools to access their accounts,” the spokesperson said. “We recently updated our Online User Agreement to provide greater clarity about those risks and to reinforce that customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents.”
Arizona opens a criminal investigation into OpenAI
The same week, the fight over AI agents and banks moved from fine print to a criminal probe.
Arizona Attorney General Kris Mayes said Wednesday that her office has opened a criminal investigation into OpenAI and other AI companies over the “apparent infiltration of banks by rogue AI agents.” She announced it at a press conference in Chicago.
Mayes declined to name the other companies under investigation, and she wouldn’t name the Arizona banks her office has contacted. She said investigators are examining whether Arizona state law is being violated on an ongoing basis.
“The state of Arizona is very worried about this, and we are looking at it from multiple angles,” Mayes said.
A criminal investigation doesn’t mean charges will be filed. But it’s a striking escalation. Just two days earlier, California Attorney General Rob Bonta sued OpenAI, alleging the company developed and tested advanced AI models without adequate safeguards. Arizona’s probe is the first known criminal investigation tied specifically to AI agents touching the banking system.
OpenAI hasn’t said much about the probe in public. The company did, however, design its own personal finance tool, launched earlier this month, as explicitly “read-only.” It “can’t move, manage, or invest your money,” according to its description. That’s a telling choice from a company whose models can now do all three.
The banks are moving together
Citi isn’t acting alone. JPMorgan and Bank of America, the two largest banks in the country, announced similar policies last month. PNC changed its terms on November 18, the same day TechCrunch reported Citi’s update. Citizens emailed customers a warning this week that they’re on the hook for anything their AI assistant does.
PNC’s version takes a slightly different route. It permits connections the bank specifically provides, with defined accounts, duration, and permissions. It prohibits giving an agent a password or one-time code, and it bars browser automation or screen scraping used to get around the bank’s controls. Break those rules, PNC says, and the customer is responsible for the resulting access and activity to the extent the law permits.
The pushback isn’t limited to banks, either. United Airlines changed its terms to bar unauthorized third-party AI agents from making or changing tickets.
Four of the country’s biggest banks now have AI-agent language in their terms, and the pattern is the same everywhere: the bank didn’t build the agent, the bank didn’t invite it in, and the bank won’t pay for what it does.
Does the fine print hold up?
Here’s the catch the banks don’t advertise. A contract clause doesn’t automatically erase federal consumer protections.
The Consumer Financial Protection Bureau’s Regulation E guidance says an agreement cannot impose more liability for an unauthorized electronic transfer than the regulation allows. If a third-party service is hacked and a thief moves money without actual authority, that may still count as an unauthorized transfer, and the bank may still have to cover it.
Nobody knows yet, because no regulator or court has decided how the rule applies to a hijacked AI agent. If you intentionally gave an agent authority and it exceeded your instructions, the dispute gets much harder. If a hacker stole the credentials from the agent and emptied your account, your case is stronger, but the banks’ new terms are written to argue the opposite.
That gap between what the terms say and what a court might say is the whole story right now. Citi’s language is a bet that most customers will never test it.
What AI agents in your bank account mean for you
The AI industry spent this year selling agents as personal staff: tools that can check your spending, pay your bills, and manage your money while you do something else. The pitch assumed the plumbing would sort itself out. It hasn’t.
Banks see agents as unreliable actors with your password. They can’t tell a helpful agent from a hijacked one, and they can’t verify that a transfer your agent made is one you actually wanted. So they’re drawing a line in the terms of service instead of building the guardrails.
For now, the practical advice is simple. Don’t give an AI agent your banking password, passcode, or a shared browser session. Use a bank-approved connection if one exists, prefer read-only access, limit which accounts and how long the access lasts, and review your statements often.
The technology isn’t slowing down. Agents will get better at handling money, and more of them will ask for your login. But until the law catches up with the terms, the person holding the bill when one goes wrong is you.
FAQ
Can an AI agent access my bank account?
Yes, if you give it your login credentials. Some AI tools ask for your banking username and password so they can check balances, pay bills, or move money on your behalf.
Who pays if my AI agent gets hacked and money is stolen?
Under Citi’s new terms, you do. Citi says it will treat access with leaked credentials as authorized by you, though federal Regulation E may still limit what a bank can charge you for an unauthorized transfer, and no court has settled that question for AI agents yet.
Why is Arizona investigating OpenAI over AI agents and banks?
Attorney General Kris Mayes opened a criminal investigation into OpenAI and other AI companies over the apparent infiltration of banks by rogue AI agents. An investigation does not mean charges will follow.
Should I stop using AI tools with my bank account?
Don’t hand an AI agent your banking password. Use only your bank’s own tools or an approved connection, keep access read-only where possible, and check your statements often.
Sources: TechCrunch, Reuters, Unite.AI
