Citi AI agent terms now make the customer answerable for what a third-party assistant does with a Citi account. A new section of the bank's Online User Agreement treats activity by outside AI agents as authorized by the customer, including losses after a leaked login or a hack. Citi is not alone. JPMorgan, Bank of America, PNC and Citizens have taken similar steps, and the timing matters because budgeting apps and AI assistants keep asking for deeper access to checking and card accounts.
What the Citi AI agent terms say about liability
The language sits under a heading titled "Use of Third-Party Artificial Intelligence ('AI') Agents," according to TFTC, which dates it to November 2026. Other outlets refer to it as Section H. The details below come from outlets that quote the agreement, not from the full document, so check the current wording in your own copy.
Under that language, a customer who gives a third-party agent access is liable even if the agent leaks credentials or gets hacked. The liability extends to what hackers do with the account, according to Mezha and Decrypt. Put simply, a breach that starts with an outside app can become the customer's loss.
The carve-out matters as much as the rule. Citi's terms exempt AI agents that the bank provides itself, which customers can use without added liability, according to TechCity Authority. Fintech companies and third-party personal finance apps are the main targets of these clauses, while bank-branded agents are carved out, per TFTC. The effect is a clear nudge toward in-house tools and away from outside apps.
The agreement also says Citi does not have to investigate whether an agent acted within the customer's authority. It can cut off an agent's access at any time without notice. That combination gives the bank the final word on what counts as authorized, while the customer gets little visibility into the decision.
Banks say they will not reverse transactions that an automated assistant makes, whether those transactions were intentional or not, according to Mezha. That makes the decision to connect an agent one of the most consequential steps a customer can take.
Which banks have changed their terms
Citi's update surfaced this week. Several outlets cite a TechCrunch report published on Tuesday, Nov. 17, and Digital Today places PNC's change on Wednesday, Nov. 18. The reporting does not consistently pin down exact dates, so treat them as approximate. The main institutions break down this way:
- Citi: Added the AI agent section to its Online User Agreement.
- JPMorgan and Bank of America: Adopted similar policies in October, according to Decrypt and TechCity Authority.
- PNC: Changed its terms, with the update reported on Nov. 18.
- Citizens: Emailed customers warning them that they are responsible for what their AI assistant does.
Citi is the third-largest US commercial bank, according to TechRadar and The AI Insider. FourWeekMBA adds up Federal Reserve data to put JPMorgan, Bank of America, PNC and Citizens at about $9.6 trillion in combined assets. That figure is the outlet's own sum, not an official number, so use it as a rough measure of scale. Because the largest banks are setting these terms at the same time, their choices carry weight well beyond their own customers.
Who absorbs the loss when an agent is hacked
Bank terms have long assumed that customers log in and make transactions themselves. AI agents break that assumption. TechRadar frames the new clauses as an effort to settle who pays when an agent is compromised, and the answer favors the bank. If a customer authorizes a third-party agent, the bank treats the agent's actions as the customer's own authorized actions.
A Citi spokesperson, who was not named in the reporting, said customers are "responsible for activities conducted through those agents." The spokesperson also said the bank wants customers to understand the risks of sharing banking credentials with third-party tools, as relayed by TechCrunch. That is a clear message: the risk of the credential sits with the customer who shares it.
Why are banks so firm? The AI Insider reports that banks see AI agents as both a security risk and a competitive threat, because such tools could help consumers avoid fees and low-interest accounts. That makes this a business concern as much as a security one.
The security worry is not abstract. USA 24 has reported that GAO covert tests got ChatGPT, Claude, Gemini and Grok to plan attacks, and that OpenAI hired Hacktron to hunt flaws after its ChatGPT account hack. Those episodes show why a leaked login is dangerous when the software holding it can be manipulated.
What the clauses leave unanswered
Three open questions will decide how much these clauses mean in practice, and the current reporting does not settle any of them.
The first is enforceability. Coverage raises the question of whether the clauses would be enforced against consumers and how they interact with Regulation E, the federal rule that covers electronic fund transfers. Regulation E shapes how losses from unauthorized transfers are handled, so any clash would matter a great deal. No regulator's answer appears in the sources.
The second is the regulatory response. Coverage mentions lawmakers' AI-related calls, but it does not describe a bank-specific response from the CFPB, the OCC or Congress.
The third is spread. Whether Wells Fargo, Capital One or other large banks will adopt similar terms remains unconfirmed. No reporting confirms that they will.
How to protect your accounts now
The steps below are general precautions rather than bank-specific advice, but they address the most direct exposure.
- Read your bank's online user agreement and search for a section on third-party AI agents. Pay close attention to the parts on liability, account access and how the bank can end an agent's access.
- Do not give third-party agents or apps your banking login. Where your bank offers its own AI tool, the Citi terms suggest that option carries less liability, but confirm the same at your institution.
- Review which apps and assistants can reach your accounts. Revoke connections you no longer use, and check whether any app stores your credentials.
- Turn on transaction alerts for checking, savings and credit card accounts so that unusual activity reaches you quickly.
Small business owners should check whether their business-account terms differ. One outlet notes that the coverage concerns personal accounts, so business customers should not assume the same language applies to them.
What to watch next
Three signals will show where this goes. The first is whether the Citi language appears in primary form, which would let customers and lawyers read exactly what the bank promises and what it disclaims. The second is whether regulators weigh in on how the clauses sit alongside Regulation E. The third is whether fintech companies answer with liability promises of their own, which would show whether the banks are setting a market standard or drawing a line.
The banks have made a clear bet that customers will accept the risk in exchange for convenience. If customers push back through complaints, lawsuits or switching accounts, the terms could change. If they do not, expect similar wording to appear at more banks as AI assistants become a common way to manage money.




