Citi changed its Online User Agreement in the week of 16 November to make customers responsible for activity by third-party AI agents they allow into their accounts, joining two banks that announced similar policies in October, TechCrunch reported on 18 November. Citi’s update also assigns customers responsibility for account activity by hackers if an AI tool leaks their credentials or is itself hacked.
Key points
- Citi’s agreement places responsibility for third-party AI agent activity, including activity after a credential leak or hack, on customers.
- PNC changed its terms on 18 November. Citizens warned customers about AI assistant activity in an email that week.
- JPMorgan and Bank of America announced similar policies in October.
- Citi says customers can use AI agents it provides without additional liability under the terms.
Citi’s agreement covers agents and account hacks
AI agents used for personal finance can be given credentials for bank accounts and other financial portals. Such tools can analyse spending and investments, offer advice, or carry out transactions and bill payments. Citi’s change concerns customers who choose to give third-party agents access to their accounts.
A Citi spokesperson told TechCrunch that the bank updated its Online User Agreement to clarify the risks of sharing banking credentials or authorising access. Customers who grant that access are responsible for activities conducted through the agents, the spokesperson said.
The agreement’s treatment of a compromised tool goes beyond a payment an assistant makes as instructed. According to TechCrunch, Citi tells customers they remain responsible for what hackers do with an account if the tool leaks credentials or is hacked. Financial institutions are also telling customers they will not be responsible for reversing transactions made by automated assistants, whether those transactions were intentional or accidental.
PNC and Citizens follow October policies
PNC changed its terms on 18 November. Citizens emailed customers during the week of 16 November to warn that they would be responsible for anything their AI assistant does, according to an email TechCrunch reported seeing. Those steps followed similar policies announced by JPMorgan and Bank of America in October.
The banks’ measures take different reported forms: changes to Citi’s agreement and PNC’s terms, earlier policy announcements from JPMorgan and Bank of America, and an email from Citizens. Each addresses activity carried out through a customer’s use of an AI assistant or agent.
Anthropic took a narrower approach with the personal finance tool it launched earlier in November. The company bills it as “read-only” and says it “can’t move, manage, or invest your money.” Its stated limits leave out the transactions and payments that some other agents can handle.
Citi executive favours agent-company liability for travel
At the Skift Commerce Summit in New York on 11 November, Abhinav Anand, Citi’s head of value cards, lending and commerce for U.S. consumer cards, addressed who should pay when an AI agent books the wrong trip, Skift reported. Anand said the question of agent liability was one banks, payment networks, merchants and AI companies were trying to answer.
For an incorrect travel booking, Anand said Citi’s view was that liability should fall “to the agent themselves, the company that is actually coming with the agent”, rather than the hotel, airline or card issuer. Skift outlined one possible arrangement in which agents seeking to transact on Visa or Mastercard might have to accept some liability.
Citi’s account-access terms address a different transaction: a customer authorising a third-party tool to use a bank account. Citi’s terms also say customers can use AI agents the bank provides without taking on additional liability.