New on GitHub
Sources 33Updates always the freshest
0 Newsletter
Animated illustration for: Citi: if an AI assistant moves your money, the bank says it’s on youHCitiJPMorganBofAPNCCitizens2026-11
AI News · Nov 19, 2026 · 4 min

Citi: if an AI assistant moves your money, the bank says it’s on you

Thinking of giving an AI assistant your bank login? I show what Citi added to its terms, who pays for a mistake or a hack, and how to use AI with money safely.

In short
  • In November Citi added Section H on third-party AI agents to its online banking agreement: anything an agent does with your login counts as authorized by you.
  • That includes credentials leaked from the AI tool. TechCrunch reports JPMorgan, Bank of America and PNC have similar rules.

Citi, the third-largest commercial bank in the United States, added a new section on third-party AI assistants to its online banking agreement in November. The gist is simple: if you give such a tool your username and password, the bank will treat everything it does as your decision. That holds when the tool makes a mistake, does something you never asked for, or gets hacked. I read the whole section, because it is the clearest signal yet of who pays when AI assistants handle money.

What Section H actually says

The new section is titled Use of Third-Party Artificial Intelligence Agents. The bank defines an AI agent broadly: any software that acts on your behalf or at your instruction, whether autonomously, semi-autonomously or not autonomously at all. Then come four sentences worth knowing:

  • if you give an agent your login details, every access and every action by that agent is treated as authorized by you,
  • if those details leak from the tool to other people or other agents, their actions are treated as yours too,
  • the bank has no obligation to check whether an action matches the prompt you gave the agent, even when it contradicts it,
  • it is your job, not Citi’s, to monitor whether the agent acts accurately and within the authority you gave it.

There is one carve-out: you are responsible for the agent’s errors except as otherwise permitted under applicable law. That exception matters, but the agreement does not say when it applies, and I am not going to guess. The bank also reserves the right to block any third-party agent at any time without notice. Finally, it reminds you that the Citi privacy notice does not cover the AI tool you let in.

The leak trap

The strongest part is the one about leaks. Your online credentials no longer live only in your head and on your phone, but also on the servers of whichever company built the assistant. If that company is breached and someone drains your account, Citi, according to this text, will treat those transactions as authorized by you. In practice, your money is only as safe as the weakest company in the chain, not as safe as the bank.

The agreement explicitly excludes AI tools that Citi itself provides in its own app. In my view that is no accident: the bank keeps a comfortable path open for its own assistants while putting a high risk threshold in front of everyone else’s.

Not just Citi

According to TechCrunch, JPMorgan and Bank of America announced similar policies in October, and PNC changed its terms on November 18. Citizens emailed customers with the same warning. In contrast, TechCrunch notes that the personal finance tool Anthropic launched this month is read-only and cannot move money. The market is splitting: banks want to control who touches the money, and AI companies are not fighting that for now.

How to use AI with money sensibly

If you want AI to help with your finances, a few rules follow directly from the agreement itself:

  • never paste your bank password into any AI tool, however eager it is to help with transfers,
  • for spending analysis, a statement file or read-only access through an official bank connection is enough,
  • approve every transfer yourself, in the bank’s app, not in a chat with the assistant,
  • if you are building a finance product, do not ask users for their bank login, because you are pushing all the risk onto them.

Citi’s text applies to customers in the US. I have not checked the terms of banks in other countries and I am not claiming they look the same, but the direction is clear. I write more about where a human must stay in the loop in the Human in the loop hub.

Sources

Human in the loop

Related