Citi and Other Banks: You Are Liable for What Your AI Agent Does

· Cortex Solutions

A glowing abstract AI agent reaching toward a digital padlock and bank card in front of an open vault door

Citi updated its terms this week: actions by third-party AI agents count as customer-authorized, even after a credential leak. PNC and Citizens moved too. What it means.

As technology companies race to launch AI agents that can analyze spending, pay bills and move money on a user's behalf, the response has come from the other side: the banks. According to a TechCrunch report published on November 18, 2026, Citibank changed its terms of service this week to tell customers plainly: if you give a third-party AI agent access to your account, you are responsible for everything it does, even if it gets hacked.

What do Citi's new terms say?

According to Unite.AI on November 18, Citi added a section called Section H to its Citi Online User Agreement covering third-party AI agents that access accounts through Citi Online or Citi Mobile. Its main points:

  • If a customer gives an outside agent their login credentials or otherwise permits access, the bank will treat the agent's access and actions as authorized by the customer.
  • If those credentials are later shared or leaked to other people or agents, actions taken with them are still attributed to the customer, and the bank has no obligation to investigate them, even when they conflict with the customer's instructions.
  • The customer is responsible for errors caused by the agent misinterpreting instructions or exceeding the authority granted, "except as otherwise permitted under applicable law", and it is the customer's job to monitor the agent's accuracy.
  • The bank reserves the right to limit or terminate any third-party agent's access at any time without notice.
  • Granting an agent access authorizes the bank to disclose customer information to it, including sensitive data, and Citi's privacy notice does not apply to that agent.

The section excludes agents and tools provided by Citi itself.

A Citi spokesperson told TechCrunch that, as third-party AI tools become more widely available, the bank wants customers to understand the risks of sharing banking credentials or authorizing those tools to access their accounts, and that the update reinforces that customers who choose to grant such access are responsible for activity conducted through those agents.

Not an isolated case

According to TechCrunch, the two largest US banks, JPMorgan and Bank of America, announced similar policies in October. Citizens emailed customers this week warning that they are on the hook for what their AI assistants do. PNC changed its digital agreement on Wednesday, November 18, and took a stricter approach, according to Decrypt on November 19: customers may use bank-approved connection methods that provide limited access, but they cannot give third-party AI agents their passwords or one-time security codes, and agents generally cannot initiate payments unless PNC expressly permits it.

Why now?

Decrypt reports that these moves follow a banking security scare on October 22, in which major banks detected AI agents logging into customer accounts and sometimes acting beyond user instructions, and executives struggled to determine which agents had been properly authorized. It also notes that the US Office of the Comptroller of the Currency (OCC) has issued examiner guidance on outside agents, payment controls and human approval, but that guidance is nonbinding.

TechCrunch sees another angle: banks view AI agents as a competitive threat. If agents become a main pathway for financial decisions, banks will fight to control that pathway, which is why Citi's terms spare its own agents from this burden.

Do consumer protections disappear?

Not necessarily. Decrypt explains that US federal rules generally require banks to investigate unauthorized electronic transfers and limit customers' losses from fraud, and that these protections cannot simply be waived through a banking agreement. How they apply to transfers made by an agent the customer authorized, however, remains unclear. The new legal question is whether what the agent did was "authorized" because the customer handed it the keys.

What does it mean for Saudi businesses?

The news is American, but the lesson is general. Here is Cortex's reading of it:

  • Before connecting any AI tool to a bank account, payment gateway or financial system, review the bank's or provider's terms and ask directly: who bears the error if the agent acts outside its instructions?
  • Never hand passwords or verification codes to an outside tool. Use official, approved connections and read-only permissions wherever possible, the direction PNC chose.
  • Route every payment or transfer through human approval, however reliable the agent seems.
  • Keep a clear log of everything your AI assistant does, because the sentence repeated across these terms is that monitoring the agent is the account holder's responsibility.
  • Prefer an assistant that runs inside your own systems and under your control, with defined permissions and reports on its work, over a general tool you give the keys to your accounts.

AI agents will become part of everyday work, but what Citi's move and those that followed reveal is that responsibility will not pass to the machine. It stays with whoever granted the access. If you are planning to automate your business with AI in a safe and controlled way, you can learn about Cortex's service here: https://cortex.sa/services/ai-automation

Sources: TechCrunch (November 18, 2026), Unite.AI (November 18, 2026), Decrypt (November 19, 2026).

Read next