Citi Shifts AI Agent Liability to Customers in Terms Rewrite
Citi's November 2026 user agreement treats third-party AI agent actions — even with leaked credentials — as customer-authorized, with no bank duty to investigate.
By Olivia Hart
4 min read

What's News
- Citi's November 2026 update to the Citi Online User Agreement, reported by TechCrunch on November 18, 2026, makes customers responsible for third-party AI agent activity.
- Access by agents using credentials later shared or leaked is still treated as customer-authorized; Citi has no obligation to investigate.
- The Citi Privacy Notice does not apply to third-party agents, which may receive sensitive personal information.
- The separate Agreement for Online Funds Transfers (Section 4) deems agent-initiated transfers directly authorized by the customer.
- Citi can limit, suspend, or terminate agent access at any time without notice.
By Olivia Hart
Citi will treat every action taken by a third-party AI agent in a customer's account as customer-authorized — even when the credentials that agent uses have been leaked to strangers. That is the core of the bank's November 2026 update to its Citi Online User Agreement, reported by TechCrunch on November 18, 2026.
The new Section H governs any third-party AI agent or agentic AI that accesses accounts or digital services through Citi Online or Citi Mobile. Citi defines an AI agent broadly: software or other technology that takes autonomous, semi-autonomous, or non-autonomous action for a customer or someone acting on the customer's behalf. Agents, tools, functions, and features that Citi itself provides through those channels are excluded from the section. In plain words, the bank's own products are carved out; everyone else's are not.
What does the customer actually sign up for?
Citi cautions customers against using third-party agents to access accounts, provide information or instructions, or conduct transactions. A customer who supplies online credentials or otherwise permits access accepts that Citi will treat the agent's access and actions as authorized.
The treatment extends well beyond the original agent. If credentials handed to an agent are subsequently shared, leaked, or otherwise distributed to other people or AI agents, access and actions using those credentials are also treated as customer-authorized. Citi states it has no responsibility or obligation to investigate the authority or validity of those actions — even when they conflict with the customer's prompts or other instructions. Notably, the terms do not describe any mechanism for a customer to prove credentials were stolen rather than negligently shared.
The customer also absorbs the consequences of agent failure. The agreement assigns responsibility to the customer for actions or errors resulting from an agent's incorrect interpretation or execution of instructions, unintended actions, and actions exceeding the granted authority, except as otherwise permitted under applicable law. Monitoring the agent's activities for accuracy, completeness, and compliance with the granted authority is likewise the customer's job, not the bank's. The takeaway: once access is granted, the operational risk sits entirely on the account holder's side of the ledger.
Who controls access — and who sees the data?
Citi reserves the right to limit, prevent, suspend, or terminate access by any third-party AI agent at any time, without notice. The bank can restrict an agent's route into its digital services. The contractual responsibility for activity already permitted, however, remains with the customer. The asymmetry is hard to miss: Citi can cut an agent off unilaterally, but the customer cannot shed liability for what the agent did beforehand.
The data provisions go further. Granting an agent account access also authorizes Citi to disclose customer information — including sensitive personal information — to that agent. The Citi Privacy Notice does not apply to the third-party agent, and Citi states it does not control how the agent collects, uses, stores, protects, or shares customer data. In effect, the access decision and the data-exposure decision are the same decision, and the privacy safeguards customers might assume apply do not.
Do money transfers get different rules?
No. Citibank's separate Agreement for Online Funds Transfers carries its own November 2026 update notice addressing third-party AI agents. Its Section 4 makes customers solely responsible for instructions supplied to Citi by an agent acting on their behalf.
Agent-initiated transfers are deemed initiated and directly authorized by the customer, without further investigation by Citi. That holds even when the instructions exceed or conflict with the prompts or other actions the customer actually took with the agent. The bank is saying, in contractual language, that a prompt is not the same thing as authority — the authority comes from opening the door.
What does this mean for personal-finance AI?
For personal-finance AI providers, the result is a demanding accountability model. Their products can reach Citi accounts, but the bank can terminate that access at any time without notice, and customers retain responsibility for monitoring agent conduct and for transfers the bank treats as directly authorized.
The agreements collectively draw a line between a customer's intent and the legal authority created by granting account access. What remains unsaid is equally telling: the terms offer no described dispute process for customers who believe an agent acted outside their wishes. For AI startups building on banking access, Citi's terms set a precedent worth watching — and one that shifts the burden of proof toward the customer.
Source: unite.ai
More from Olivia Hart
Show full bio
Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,378 articles