News

Citi Makes Customers Responsible for Third-Party AI Agents

Agentic Commerce Atlas · · Updated

Good for: Leaders · Builders

Citi has written AI agents into its online banking terms and put the risk of using an outside one on the customer. In November 2026 Citi added a new Section H, “Use of Third-Party Artificial Intelligence (“AI”) Agents,” to its Citi Online User Agreement. It cautions customers against letting a third-party AI agent access their accounts through Citi Online or Citi Mobile and says that if they do, “Citi will consider any access and actions taken by the agent as authorized by You.” TechCrunch reported on 18 November that PNC changed its terms the same day and that Citizens emailed customers a similar warning that week.

What Citi’s new Section H says

The section applies to a customer’s use of “a third-party AI agent or agentic artificial intelligence to access your accounts or use digital services through Citi Online or Citi Mobile.” It defines an AI agent as software or other technology that takes “autonomous, semi-autonomous, or non-autonomous action” for the customer. Its main terms:

  • Actions count as the customer’s. Citi treats any access and actions taken by the agent as authorized by the customer, and says it has no responsibility to investigate the authority or validity of those actions.
  • Monitoring is the customer’s job. “You agree it is your responsibility and not Citi’s to monitor the activities of the third-party AI Agent,” the terms say.
  • Agent errors and leaked logins. The customer is responsible for actions or errors that result from the agent’s incorrect interpretation of instructions. If the online credentials given to the agent are further shared, leaked or otherwise distributed to other persons or AI agents, Citi will also treat that access as authorized by the customer.
  • A legal qualifier. Responsibility for agent actions that exceed the authority the customer granted applies “except as otherwise permitted under applicable law.”
  • Citi can shut agents out. Citi reserves the right to limit, prevent, suspend or terminate access by any third-party AI agent at any time without notice.
  • Citi’s own tools are excluded. The terms do not apply to agents, tools, functions or features that Citi itself provides through Citi Online or Citi Mobile.

A notice at the top of the agreement says it was last updated in November 2026 to add the section, and that the new terms “caution against your use of third-party AI agents” and set out the customer’s “responsibility for the actions of third-party AI agents acting on your behalf.” A Citi spokesperson told TechCrunch the agreement makes customers “responsible for activities conducted through those agents.”

What PNC and Citizens did, according to TechCrunch

TechCrunch reported that PNC changed its terms on 18 November and that Citizens emailed its customers that week warning that they are on the hook for anything their AI assistant does. It described both as moving in the same direction as Citi, and said JPMorgan Chase and Bank of America announced similar policies in October. PNC’s and Citizens’ own wording was not available to the Atlas, so the details of their terms are attributed to TechCrunch’s report. JPMorgan’s position, which blocks outside agents from acting for retail customers, is covered in the earlier News report.

Why Citi’s terms matter for agent payments

The question of who pays when an agent gets it wrong has so far been answered mostly in principle. Six banks’ September agentic commerce principles say liability should reflect where risks and errors are introduced, and JPMorgan’s chief executive spoke of a “liability transfer” without setting out terms. Citi has now put an answer in a customer contract: for an outside agent that logs in to online banking, the customer carries the risk, largely without exception, and the bank keeps the right to block the agent.

Section H is written around agents that use a customer’s online banking access. It does not mention card payments, payment tokens or merchants. Agent payment schemes built on network tokens and explicit user approval, such as ChatGPT Wallet or the network programs described in AI shopping agent payment authorization, work differently from an agent holding a customer’s banking login. For companies building personal agents that pay bills or move money from a bank account, the terms mean that a Citi customer who shares online banking access with such an agent carries the risk of what it does.

What the banks have not said

Citi’s terms do not say whether any outside agent, or any way of identifying an agent such as a protocol, would be treated differently, and they name no agents. The Atlas found no list from Citi of agents it allows or blocks. PNC’s and Citizens’ terms have not been published in a form the Atlas could check. Citi’s terms do not say how they interact with existing consumer protections, beyond an “applicable law” qualifier on actions that exceed the customer’s authority, and TechCrunch’s report does not say whether PNC or Citizens addressed it.

See also agentic commerce risks and readiness and the agentic commerce timeline.

Frequently asked questions

What did Citi change about AI agents?

Citi updated its Citi Online User Agreement in November 2026 to add a new Section H, 'Use of Third-Party Artificial Intelligence (AI) Agents.' It cautions customers against using third-party AI agents to access their accounts or digital services through Citi Online or Citi Mobile and says that if they do, 'Citi will consider any access and actions taken by the agent as authorized by You.'

Who is responsible if a third-party AI agent makes a mistake with a Citi account?

Under the new section the customer is. The terms say monitoring the agent is 'your responsibility and not Citi's,' that the customer is responsible for actions or errors that result from the agent's incorrect interpretation of instructions, and that if the login credentials given to the agent are shared or leaked to other persons or agents, Citi will also treat that access as authorized by the customer. Responsibility for agent actions that exceed the authority the customer granted applies 'except as otherwise permitted under applicable law.'

Do the terms cover Citi's own AI tools?

No. The section says it does not apply to AI agents, tools, functions or features that Citi itself provides through Citi Online or Citi Mobile.

Which other banks have made similar changes?

TechCrunch reported on 18 November 2026 that PNC changed its terms that day and that Citizens emailed customers that week warning they are on the hook for anything their AI assistant does. TechCrunch also said JPMorgan Chase and Bank of America announced similar policies in October. The PNC and Citizens wording has not been published in a form the Atlas could check.

Primary sources

  1. Citi® Online User Agreement · Citi, 2026-11
  2. More banks say customers using AI agents are on their own, even if they get hacked · TechCrunch, 2026-11-18