Citi Now Treats Anything Your Third-Party AI Agent Does in Your Account as Authorized
Citi added Section H to its Online User Agreement in November 2026: third-party AI agent actions, including after a credential leak, are treated as authorized by the customer.
Citi just wrote the liability shift into the contract. Its Online User Agreement, updated in November 2026, adds a Section H on third-party AI agents. If you hand your Citi Online or Citi Mobile login to an outside agent, or otherwise let it into the account, Citi will treat every action that agent takes as yours. If those credentials are then leaked, shared or passed to another person or another agent, Citi still treats whatever happens next as authorized.
The bank says it has no duty to investigate whether those actions matched your prompts. You are on the hook for a wrong interpretation, an unintended transfer, or an agent that exceeds the authority you thought you gave it, "except as otherwise permitted under applicable law." Monitoring the agent is your job, not Citi's. The bank can block any third-party agent at any time without notice.
The same clause carves out Citi's own tools. Agents, features or functions Citi provides through Citi Online or Citi Mobile are not covered by Section H. Outside agents are. By letting one in, you also authorize Citi to disclose your information, including sensitive personal information, to that agent. Citi's Privacy Notice, the agreement says, does not apply to it.
The quote that matches the PDF
A Citi spokesperson told TechCrunch the bank wanted customers to understand the risk of sharing credentials or authorizing third-party AI tools, and that the update "reinforce[s] that customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents." That is the same rule as the printable terms, in plainer English.
The definition in Section H is broad on purpose. An "AI Agent" is software that takes autonomous, semi-autonomous or non-autonomous action on your behalf, or on the instruction of anyone acting for you. That covers a frontier chatbot with your password, a personal-finance copilot, and a script that only clicks through a bill pay flow.
The rest of the industry already started
Citi did not invent the category. Finpresso covered JPMorgan when Jamie Dimon said the bank would not let "just any outside agent" inside, and CFO Jeremy Barnum said it was already blocking what it could. Cyberpresso reported the wider US bank reaction as Morgan Stanley moved to block outside agents and others talked about read-only access, the model they already use for budgeting apps.
Citi's November language goes a step further than a block. It keeps the door technically open, warns you not to walk through it, and then says that if you do, a hack of the tool you trusted still looks, to Citi, like you.
What the terms do not do is repeal "applicable law." Regulation E and other rules still sit outside this PDF. Citi is telling customers how it will treat the session inside its own systems. Whether a court or an agency agrees, after a real loss, is a fight the clause is written to win on paper.
Some offers on this page may be paid placements or contain affiliate links.
Finpresso: daily AI & finance brief
Free daily newsletter, read in 5 minutes.
Subscribe free