Citi Puts AI Agent Risk on Customers as Banks Shift Liability
Citi says customers who let outside AI agents into their accounts are responsible for what happens, even if the tools are hacked.
Citigroup has rewritten its online banking terms so that customers who give third-party artificial intelligence agents access to their accounts are responsible for whatever those tools do, including losses if the agent is hacked, TechCrunch reported on Wednesday. Citi is the latest in a growing group of large U.S. banks telling customers that, when they hand an AI assistant the keys to their money, they carry the risk.
What happened
According to TechCrunch, Citi, the third-largest commercial bank in the United States, changed its Online User Agreement this week. Under the new language, customers who connect an outside AI agent to their account are on the hook for activity carried out through it. If the tool leaks login credentials or is breached, the customer is responsible for anything an attacker does with the account. The terms carve out an exception for AI agents that Citi provides itself, which customers can use without taking on that added liability.
A Citi spokesperson told TechCrunch the bank wants customers to understand the risks of sharing banking credentials or authorizing third-party tools, and said the agreement was updated “to provide greater clarity about those risks.”
Citi is not alone. TechCrunch reported that Citizens Financial Group emailed customers this week warning them they are liable for anything their AI assistant does, and that PNC, the eighth-largest U.S. bank, changed its terms on Wednesday. JPMorgan Chase and Bank of America, the country’s two largest banks, announced similar policies in October.
Why banks are pulling back
AI agents are software tools, offered by major AI labs or by specialized finance apps, that can act on a user’s behalf. Given bank or brokerage logins, they can analyze spending, suggest investments, pay bills and move money. Banks have said they will not take responsibility for reversing transactions that automated assistants make, whether intended or not.
Bank executives have framed the issue mainly as one of security and authorization. Jamie Dimon, chief executive of JPMorgan Chase, said on the bank’s third-quarter earnings call that it would limit AI agents’ access to client accounts because “you don’t want rogue agents and there could be thousands of them,” Retail Banker International reported. Brian Moynihan, chief executive of Bank of America, told CNBC the bank believes “there should be a human in the loop” to approve money leaving a customer’s account, and said it is developing protocols to decide who is liable for AI-authorized payments.
At the Money20/20 fintech conference in Las Vegas in October, Naveed Anwar, Citi’s global head of AI, digital and data platforms in its services business, said the bank would not rush to let outside agents connect. “There are places in which we want to go faster. This is not one of those places,” he said, according to Banking Dive. Anwar proposed a “know your agent” process, modeled on the know-your-customer identity checks banks already run, to verify agents before granting access. Bill Demchak, chief executive of PNC, argued that lenders should add friction rather than remove it when agents are involved, Banking Dive reported.
There is also a competitive dimension. Retail Banker International, citing the Financial Times, Bloomberg and CNBC, reported that banks worry agents could steer customers’ idle cash out of low-interest deposit accounts and into higher-yielding alternatives, eroding a key source of bank profits. TechCrunch noted that markets briefly fell on the prospect that AI could help consumers dodge fees and find better rates, and argued that banks have an incentive to control how agents reach their customers.
Background
The pushback accelerated in October. Reuters reported at the time that Citi, JPMorgan and Bank of America were blocking external agents such as Muse, a personal assistant from Meta Platforms, from accessing bank accounts, according to Banking Dive. Charles Schwab chief executive Rick Wurster told Bloomberg the brokerage was blocking unauthorized agents, including Muse, that tried to move client money. Citi said then that it does not permit third-party agents to log into customer accounts, though customers can still grant limited, read-only access to certain approved open banking providers.
Regulation has not caught up. Retail Banker International reported that Wall Street executives have contacted Federal Reserve officials about the issue and that regulators have privately supported the banks’ moves, while officials believe the law may need to spell out how customers authorize autonomous software and who bears the loss when something goes wrong.
What comes next
Some AI companies are designing around the problem. TechCrunch noted that Anthropic’s personal finance tool, launched earlier this month, is billed as read-only and says it “can’t move, manage, or invest your money.”
For now, the key points for consumers using these tools are:
- Several major banks now say losses from outside AI agents, including hacks, fall on the customer.
- Agents offered by the banks themselves are treated differently under Citi’s terms.
- No federal rule yet assigns liability for transactions made by AI agents.
How that liability question is settled, by regulators, courts or the banks’ own contracts, is likely to shape whether AI agents become a mainstream way to manage money or stay confined to read-only roles.


