Read Section H before you read the headline

The document is short enough to check yourself, and it is worth doing, because the coverage so far compresses it into "banks say agents are your problem." Section H of Citi's Online User Agreement, added in the agreement's November 2026 update, starts with a caution: Citi cautions customers against using a third-party AI agent to access their accounts, give Citi information or instructions, or perform transactions through Citi Online or Citi Mobile. It then sets out what follows if a customer disregards that caution.

If you hand an agent your online credentials, or otherwise let it in, Citi "will consider any access and actions taken by the agent as authorized by You." The bank states it has no responsibility to investigate whether those actions were valid, even when they conflict with the prompts or instructions you gave the agent. Errors from the agent misreading your instructions, actions you never intended, and actions beyond the authority you granted are yours, with one qualifier the agreement itself carries: "except as otherwise permitted under applicable law." Monitoring the agent, to check its work stays accurate and inside the authority you gave it, is also assigned to you.

Citi reserves a right for itself in the same section. It can limit, prevent, suspend or terminate any third-party agent's access at any time, without notice.

The leak clause, in Citi's own words

Treating an agent's actions as the customer's is now common drafting. JPMorgan and Bank of America announced similar policies in October 2026, and PNC changed its terms on November 18, 2026, the day TechCrunch reported Citi's update. Citizens emailed its customers a similar warning this week, according to an email TechCrunch saw. Whether those banks' texts contain a clause like Citi's leak clause is not established by the reporting; what follows is Citi's language, verified in Citi's agreement.

If the credentials you gave an agent are "further shared, leaked, or otherwise distributed to other persons or AI agents," then access and actions taken with those credentials are also treated as authorized by you. Read that slowly. The clause does not require the agent you chose to have caused the leak, and it does not stop at the agent misbehaving. It covers a second party you never chose, using credentials that left the first party's control, doing things that may have nothing to do with your instructions. Under the agreement's framing, Citi will treat that as your authorized activity, and the bank states it has no duty to investigate it.

A separate Citi document points the same way for money movement, per Unite.AI's account of it. Unite.AI reports that Citi's Agreement for Online Funds Transfers, which carries its own November 2026 update notice, says in its Section 4 that customers are solely responsible for instructions an agent supplies on their behalf, with agent-initiated transfers treated as initiated and directly authorized by the customer. ALFW has not read that second agreement first-hand; the Online User Agreement above is the one verified here.

What the terms do not settle

Two carve-outs matter. First, Section H does not apply to agents, tools or features Citi itself provides through its own channels. The section draws its line at who provides the tool, not at the technology. A Citi spokesperson told TechCrunch the update was made "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."

Second, a contract clause is a position, not a ruling. The agreement's own "except as otherwise permitted under applicable law" line concedes that some of this may not hold where consumer-protection law says otherwise, and no source reporting this week's changes identified a court decision or a regulator statement testing these clauses against a real loss. What the text settles is how Citi says it will treat these actions; how a dispute over an agent-made transaction would end is a separate question nobody has answered yet.

The design contrast is already on the market. Anthropic's personal finance tool, launched earlier this month, is billed as read-only: it can look at accounts and cannot move, manage or invest money. Read-only access limits what a tool can do with money. It does not put a tool outside Section H's scope — the section covers customers who provide credentials or otherwise permit access — and an agent that can read an account can still expose what it reads. The difference the terms create is about transaction authority: an agent that holds your password and can transact sits in a different position from one that can only look, and Section H assigns the customer the monitoring duty either way.

Sources

  1. [1] Citibank, "Citi Online User Agreement," last updated November 2026 (header notice and Section H, "Use of Third-Party Artificial Intelligence ('AI') Agents"), read first-hand in fullRead source
  2. [2] TechCrunch, Tim Fernholz, "More banks say customers using AI agents are on their own — even if they get hacked," November 18, 2026Read source
  3. [3] Unite.AI, "Citi Makes Customers Responsible for Third-Party AI Agent Activity," November 18, 2026 (the funds-transfer agreement detail is from this account; ALFW has not read that second agreement first-hand)Read source