Citi's online banking terms now carry a plain message about outside AI agents: if you hand an agent your login, Citi will treat what the agent does as authorized by you, and the same goes for anyone who gets that login from the agent.

The new Section H of the Citi Online User Agreement ↗, which the agreement says was added in November 2026, opens with a warning: "You are cautioned against using any third-party AI Agent to access your accounts or use any digital services." It says an agent is software that "takes autonomous, semi-autonomous, or non-autonomous action on behalf of, or at the instruction of, You." Archived copies of the agreement show no agent section in December 2025 and Section H in place by Nov. 7, 2026.

If a customer goes ahead anyway, the terms say "Citi will consider any access and actions taken by the agent as authorized by You." The section then states: "If the Online Credentials that you provided to the third-party AI Agent are further shared, leaked, or otherwise distributed to other persons or AI agents, Citi will consider any access and actions taken with the Online Credentials as authorized by You."

Customers are also responsible, the terms say, for an agent's "incorrect interpretation or execution of your instructions, unintended actions performed by the third-party AI Agent, or actions taken by the third-party AI Agent that exceed the authority granted by you," a sentence that ends "except as otherwise permitted under applicable law." Monitoring the agent is the customer's job, and Citi can shut out any outside agent without notice. Citi's own agent tools are exempt.

A Citi spokesperson told TechCrunch on Nov. 18 ↗ that the update was meant "to reinforce that customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents." TechCrunch summarized the leak clause as leaving customers on the hook even if the AI tool is hacked. Neither describes an incident involving a Citi customer.

Not only Citi

Chase's Digital Services Agreement ↗ says that, effective Oct. 27, outside parties "including External AI" may not use a customer's login credentials, while approved API access to account data is allowed. A customer who lets an outside party in anyway, the agreement says, is deemed to "expressly authorize that third party to transact in your account" and agrees "you are responsible for any resulting errors." The same section says those consequences "apply equally to" any "account data access you authorize through APIs," so approved connections carry the same customer-responsibility terms.

PNC's updated Digital Services Agreement ↗, which takes effect Dec. 19, tells customers not to give an agent their user ID, password or one-time passcodes. Unless PNC permits it for a particular service, agents may not make payments or transfers, and customers who break those rules answer for the results as far as law allows, the agreement says.

Under Wells Fargo's revised online access terms ↗, which start Nov. 24, instructions from an agent you let in are, the bank says, "deemed to have been initiated and authorized directly by you," unless law provides otherwise. We noted that clause in our earlier piece on Barr's bank-run warning ↗.

Bank of America emailed customers that if they share their login, "you're responsible for activity resulting from that access," TechCrunch reported in October ↗, and a bank spokesperson confirmed the email. TechCrunch also reported Nov. 18 that Citizens sent a similar email that week; AINewsIQ has not seen it.

None of these agreements bans agents outright; Chase and PNC point to bank-approved connections.

Fed Governor Michael Barr warned this month that agents could accelerate bank runs. In a Nov. 19 speech at the Singapore FinTech Festival ↗, Monetary Authority of Singapore managing director Chia Der Jiun said agents "have the potential to move money and allocate assets more quickly for smaller incentives." Bloomberg, reporting the central bank's warning ↗, framed it as a liquidity and stability risk. Chia also said some users already give agents their financial logins, and that he "would not advise giving personal agents such access, as there is insufficient assurance of safety."

Our analysis: who carries the risk

Our read: the key decision these terms leave to the customer is whether to give an agent a password at all. Citi also says it has "no responsibility or other obligation to investigate the authority or validity" of such actions, even when they conflict with what the customer told the agent to do.

Citi, Chase, PNC and Wells Fargo each include an applicable-law carve-out without saying how it would apply in a given dispute. Still, the default is that the customer answers for the agent, which could push agent makers to design for safety.

What builders and users should do

Our suggestions, not bank guidance:

Users - Don't give an agent your bank password or one-time codes. The agreements warn that sharing credentials can cause resulting agent activity to be treated as authorized, subject to applicable law. - Prefer read-only or bank-approved connections. If you already shared a login, change it; Chase's terms require that. - Check activity often and revoke agent access you no longer use. PNC says you stay responsible until a revocation takes effect.

Builders - Don't collect bank credentials: under Citi's terms, a leak from your service lands on your users. Use the bank's approved connection, and start read-only. - Require a person to confirm every payment or transfer. - Log each action with the instruction behind it.

What to watch

  • Whether banks open an approved route for outside agents to make payments, not just read data.
  • Whether regulators address how terms like these sit alongside federal rules on unauthorized electronic transfers.
  • Wells Fargo's terms take effect Nov. 24 and PNC's on Dec. 19.
  • Whether more banks turn email warnings into written terms.

Evidence note: Citi, Chase, PNC and Wells Fargo terms are quoted from each bank's published agreement; archived copies of Citi's agreement are Internet Archive captures from [Dec. 11, 2025](https://web.archive.org/web/20251211141018/https://www.citi.com/tandc/view-printable-version) and [Nov. 7, 2026](https://web.archive.org/web/20261107105444/https://www.citi.com/tandc/view-printable-version). Chia's words come from the speech as MAS published it. The Citi spokesperson's statement and the Bank of America and Citizens emails are as reported by TechCrunch; AINewsIQ has not seen the Citizens email. AINewsIQ has not read Bloomberg's full article, only its post on X.

This article was researched and drafted with AI assistance, then fact-checked and edited by Paul Woll, who is responsible for its content. Sources are linked inline.