NewsFor: Legal and compliance · Finance and accounting · Customer support

Citi and PNC tell customers they are liable for what AI agents do

Five large US banks now hold customers responsible for AI tools they let in, hacks included. Finance should ban shared bank logins; legal should set a rule.

Citi changed its Online User Agreement this week so that customers who give a third-party AI agent access to their accounts "are responsible for activities conducted through those agents", a Citi spokesperson told [TechCrunch](https://techcrunch.com/2026/11/18/more-banks-says-customers-using-ai-agents-are-on-their-own-even-if-they-get-hacked/) on 18 November 2026. A third-party AI agent here is an assistant from another company that logs in to the bank for the customer, usually with the customer's own username and password, to read spending or to pay bills. TechCrunch reports the update goes one step further than a warning: if the AI tool leaks the login or is hacked, the customer answers for whatever the hackers do with the account.

Five banks, one direction

  • Citi: agreement updated in the week of 16 November. TechCrunch reports its terms say AI agents that Citi itself provides carry no added liability for the customer.
  • PNC: changed its terms on 18 November, TechCrunch reports, without quoting them.
  • Citizens: emailed customers this week that they are responsible for anything their AI assistant does, according to an email TechCrunch saw.
  • JPMorgan and Bank of America: "similar policies last month", TechCrunch reports. Bank of America's email told customers never to share a login with "an Al-enabled tool or agent that offers to act on your behalf", and added: "you're responsible for activity resulting from that access", as [TechCrunch quoted it on 27 October](https://techcrunch.com/2026/10/27/bank-of-america-kindly-asks-that-you-dont-share-your-banking-password-with-your-ai-assistant/).

One AI vendor has built around the problem. Anthropic's Personal Finance tool for Claude, launched on 5 November, connects to bank and card accounts through the data provider Plaid instead of a shared password, and Anthropic says access is read-only: the tool cannot move, manage or invest money, [TechCrunch reported](https://techcrunch.com/2026/11/05/claude-can-now-analyze-your-spending-with-its-new-personal-finance-tool/).

What the reports do not say

Every document TechCrunch describes is a consumer one. Neither report says whether business or commercial accounts carry the same wording, how the new terms sit with US consumer-protection rules on unauthorised transfers, or how a bank will tell an agent from the customer. No bank gave a figure for agent logins, disputed transactions or losses. Nothing is said about what branch or contact-centre staff tell a customer whose assistant moved money.

The same question in your vendor contracts

A 17 November advisory from the law firm Mayer Brown puts the question to companies that deploy agents themselves, as [PYMNTS reports it](https://www.pymnts.com/news/artificial-intelligence/2026/when-ai-agents-break-rules-who-pays/): regulators are expected to look first to the company that deployed the agent, and whether it recovers losses from its AI supplier depends on what the contract assigns and what the company can prove about how the agent acted. PYMNTS says the advisory recommends agreeing before rollout who runs each safeguard, who can suspend the agent, which records are kept (instructions, configuration changes, approvals, transactions) and how fast they can be retrieved, and walking through a simulated incident with the supplier.

What to do about bank logins and agent liability this week

  • Finance: tell the team in writing that no bank, card or payment-portal login is given to an AI tool, and ask your bank's relationship manager whether your business account agreement now carries the same clause as the consumer one.
  • Legal and compliance: if customers hold accounts with you, decide whether your terms say who is responsible when a customer's AI assistant acts on the account, and whether tools you provide are treated differently, as Citi's are.
  • Whoever signs AI vendor contracts: before the next agent goes live, add to the contract who runs each safeguard, who can switch the agent off, and which logs the supplier must hand over and how quickly.
Was this useful for your work?

Take this to work

Forward this

Citi and PNC changed their customer terms in the week of 16 November 2026 so that account holders are responsible for what third-party AI agents do in their accounts, including after a hack of the AI tool, TechCrunch reports. Nobody here should give a bank or card login to an AI tool, and our customer terms and AI vendor contracts should say who pays when an agent acts wrongly.

Questions to ask

  1. Does our business or commercial account agreement carry the same clause as the consumer one, and since when?
  2. Which staff have already given an AI tool the login to a bank, card or payment portal?
  3. If a tool needs our transactions, can it connect read-only through a data provider instead of a shared password?
  4. Do our own customer terms say who is responsible when a customer's AI assistant acts on their account?
  5. In each AI vendor contract, who runs each safeguard, who can suspend the agent, and which logs can we get and how fast?

Questions this article answers

Is a Citi customer liable if an AI agent they authorised is hacked?

TechCrunch reports that Citi's updated Online User Agreement makes customers responsible for activity through third-party AI agents they grant access to, including what hackers do if the tool leaks their login or is hacked. Citi's terms say agents Citi itself provides carry no added liability.

Which US banks hold customers responsible for AI agents on their accounts?

As of 18 November 2026 TechCrunch names five: Citi, PNC and Citizens changed terms or warned customers that week, and JPMorgan and Bank of America set similar policies in October 2026.

Do the banks' AI agent terms apply to business accounts?

The reports do not say. The documents TechCrunch describes are consumer agreements and customer emails, so a company should ask its bank whether its business account agreement carries the same clause.

Who is liable when a company's own AI agent breaks a rule?

A 17 November 2026 Mayer Brown advisory, as PYMNTS reports it, says regulators are expected to look first to the company that deployed the agent, and that recovering losses from the AI supplier depends on the contract and on records of how the agent acted.

Send this to someone who runs a teamLinkedInXBlueskyHNEmail