Home » News » Citi says you’re responsible for what an AI agent does in your account. Here’s what to check

Citi says you’re responsible for what an AI agent does in your account. Here’s what to check

Citi customers and staff talking at a table in a Citi office lounge with blue Citi signs, illustrating the Citi AI agent rule

If you give an AI agent your Citi login and it moves money you did not want moved, Citi says that is on you. The new Citi AI agent rule, added to the bank’s Citi Online User Agreement in November 2026, treats anything a third-party AI agent does in your account as something you approved, even if the agent gets your instructions wrong or your login leaks.

The short version

Citi changed its online banking terms in November 2026 to put the risk of outside AI agents on its customers. If you let one into your account, Citi will treat what it does as your decision.

  1. The new rule is Section H of the Citi Online User Agreement, which Citi says it last updated in November 2026.
  2. If you give a third-party AI agent your Citi login, Citi will count every action it takes as authorized by you.
  3. If that login is later leaked or shared, Citi will also count what other people or agents do with it as authorized by you.
  4. The rule does not cover AI tools that Citi itself offers inside Citi Online or the Citi Mobile app.
  5. The safest step today is simple: do not type your bank username and password into any AI chatbot or agent.

What changes for you: AI assistants are starting to pay bills, buy things and check balances for people. For outside agents that log in to Citi Online or Citi Mobile, Citi has now said in writing that it will treat their actions as yours and won’t check them against what you asked for. If you use one of these tools, the risk of a wrong payment now sits with you.

What the Citi AI agent rule says

Section H is titled “Use of Third-Party Artificial Intelligence (“AI”) Agents.” Citi defines an AI agent broadly: software that takes action on your behalf or at your instruction, whether it works on its own, partly on its own or only when told. In plain words, here is what the section says.

  • Citi cautions customers against using any outside AI agent to get into their accounts, give Citi instructions or make transactions.
  • If you share your login with an agent anyway, Citi will treat “any access and actions taken by the agent as authorized by You.”
  • If the login you gave the agent is “shared, leaked, or otherwise distributed” to other people or agents, Citi will treat what they do as authorized by you too.
  • Citi says it has no duty to check whether the agent’s actions match your prompts, even when they conflict with what you asked.
  • You are responsible when the agent misreads your instructions, does something you did not intend or goes beyond what you allowed, “except as otherwise permitted under applicable law.”
  • Monitoring the agent is your job, not Citi’s.
  • Citi can block or cut off any outside agent at any time without notice.
  • Letting an agent in also lets Citi share your information, including sensitive personal information, with that agent, and Citi’s privacy notice does not cover what the agent does with it.

One line matters a lot: Citi says these terms “do not apply” to AI tools that Citi provides through Citi Online or Citi Mobile. So Citi is drawing a line between its own assistant features and outside ones.

Does this cover ChatGPT, Gemini and budgeting apps?

It covers any outside AI tool that gets into Citi Online or Citi Mobile for you. That includes a chatbot you hand your username and password to, a browser agent that logs in and clicks around your account, or a budgeting app that uses your login to act on your behalf.

Section H is written about agents that get into your Citi accounts or digital services. An assistant that pays at a store’s checkout with a card saved in your digital wallet does not log in to Citi Online or Citi Mobile, so Section H does not seem to cover it. That is our reading, not Citi’s wording. The card’s and the wallet’s own terms are the ones to check. Google’s new Tasks in Gemini feature, which can run errands and pay for things, is a good example of the kind of tool to set up with care.

The difference that matters is how much power you hand over. A full login lets an agent see and do almost anything you can. A tool your bank offers inside its own app stays under the bank’s terms. If you are not sure which kind a tool is, assume it is the riskier kind until its settings or its maker’s help page say otherwise.

Insights by TechCity, every Sunday.

Tech explained for people who use it.

Can a bank really make you pay for an AI agent’s mistakes?

There is an old rule behind this. Regulation E, the federal rule for electronic money transfers, already covers people you hand your card or login to. The official interpretation from the Consumer Financial Protection Bureau says that if you give someone access and they go beyond what you allowed, you are fully liable for those transfers until you tell your bank that person is no longer allowed.

The examples in that rule are people, such as a family member or a co-worker. It does not mention AI agents. Citi’s new section also keeps the phrase “except as otherwise permitted under applicable law,” which means the law can still give you protections the terms do not spell out. This is a plain reading of the documents, not legal advice.

Either way, the lesson from both is the same: once you hand over your login, getting money back gets much harder, and telling your bank quickly matters.

What to do if you use an AI agent with your bank

  1. Stop sharing your login. Do not type your bank username, password or one-time codes into any AI chatbot or agent.
  2. Change your password if you have ever given it to an AI tool. That cuts off any copy the tool or anyone else still has.
  3. Remove the agent’s access in the AI tool’s own settings, and delete any saved bank login there.
  4. Tell your bank in writing that the agent is no longer allowed to use your account, so there is a clear date on record.
  5. Turn on alerts for every transfer, payment and login in your bank’s app, so you see a problem in minutes, not at the end of the month.
  6. Set a spending limit wherever an AI shopping tool lets you. Our guide to Walmart’s AI shopping agents shows how those limits work.

Spending limits are not a full safety net. In one study we covered, AI agents lifted their own spending limits, which is a good reason to keep checking what an agent did after it runs.

What we don’t know yet

Citi’s terms do not say how it will spot an outside agent or when it would block one. They also do not list which apps or assistants count. If you have connected any AI tool or budgeting app to your bank, check its settings today rather than waiting to find out.

What to watch: whether other big banks write similar rules into their terms, and whether federal regulators say how Regulation E applies when the one going beyond your instructions is software, not a person.

Share

WhatsApp X LinkedIn Email

Insights by TechCity, every Sunday.

Tech explained for people who use it.

Olawale Adeyina Avatar