Citi to Customers: Your AI Agent’s Losses Are Yours

Human Verified

Share to: , and

A cracked shield beside a glass bank building at dusk, with a glowing AI orb above a smartphone

Citibank has rewritten its customer terms so that anyone who lets a third-party AI agent into their bank account now carries the risk themselves. The new language covers mistakes the agent makes, transactions nobody asked for, and even losses after the AI tool is hacked.

What Citi changed

The change landed in Citi’s Online User Agreement this week. A bank spokesperson said the update gives customers greater clarity about the risks of sharing banking credentials or authorising outside AI tools to reach their accounts, and makes clear that customers are responsible for what those agents do.

Citi’s terms go further than a warning about bad transactions. If an AI tool leaks a customer’s credentials or is breached, the customer is responsible for whatever hackers do with the account. The bank will not reverse those payments either.

Citi is not acting alone. PNC, the eighth-largest bank in the country, changed its terms the same day, and Citizens emailed its customers a similar warning this week. JPMorgan and Bank of America, the two largest banks, announced similar policies last month.

The banks are reacting to a real shift in how people use AI. Agents can now be given the credentials to a bank account and other financial portals, so they can analyse spending, offer advice, and even handle transactions and bill payments on a customer’s behalf.

There is one exception in Citi’s terms. Customers who use AI agents that Citi itself provides can do so without worrying about additional liability, a carve-out that keeps the bank’s own tools on the safe side of the new rules.

Why the banks are drawing the line

Financial institutions are not thrilled with unreliable AI actors inside their systems. They are telling customers they will not be held responsible for reversing an intentional or unintentional transaction made by an automated assistant, whether the agent meant to do it or not.

The banks also see a competitive threat. AI tools that make it easier for consumers to avoid unnecessary fees or low-interest deposit accounts would eat into bank margins. Anthropic’s answer is a read-only finance tool that cannot move money at all.

The timing is no coincidence. Arizona opened a criminal investigation earlier this week into OpenAI and other AI firms over the apparent infiltration of banks by rogue AI agents, the same risk Citi’s new terms are written to guard against.

Why It Matters

Four of the biggest banks in the United States have now put the risk on the customer, and the shift lands just as AI agents move from answering questions to moving money. The liability question is no longer theoretical — it is written into the terms every customer clicks past.

The reliability gap behind the shift is documented. Visa found that AI shopping agents failed nearly half of their test runs, and the same unreliability that loses a shopping order can drain a bank account when the agent holds the credentials.

Whether the new clauses would survive a real dispute is still an open question, and the first drained account to reach a judge will test how far a bank can go in defining an agent’s actions as authorised by the customer who handed it the keys.

Background

Washington is still writing the rules for AI liability while banks write their own. The White House offered AI firms a 12-month shield for speaking up about cyber flaws, but nothing yet settles who pays when an agent moves money it should not have touched.

Source: TechCrunch’s report on the new bank terms

Share to: , and

Faheem R. Khattak

Senior automation engineer and AI marketing journo covering LLMs, algorithms, analytics, core updates, Bing search, and technical SEO.

0 Comments

Leave a Reply

Optional Take Tag:

🛡️ Spam protection: Honeypot + Time-Trap Active

This site uses Akismet to reduce spam. Learn how your comment data is processed.