Friday 20 November 2026 Edition compiled at 07.38 NL EN FR
← Tech · Other tech
Tech · Other techBanking

Citi and PNC place full responsibility for what an outside AI agent does in an account on the customer

After JPMorgan and Bank of America, three more US banks are changing their terms, and the banks' own agents are exempt from the rule.

· latest source update wo · 2 sources · Decrypt, TechCrunch · Compiled with AI
Illustration for this article

Citibank, the third-largest commercial bank in the US, changed its online banking terms this week. Anyone who gives their login details to a third-party AI agent is deemed to have approved every action of that agent themselves, even if the agent ignores instructions, exceeds its permissions or passes on the credentials, Decrypt reports. According to TechCrunch, the same applies when the credentials are stolen or the account is hacked. Agents that Citi itself offers are exempt from the rule.

PNC changed its agreement on Wednesday and distinguishes between bank-approved connections with limited access and agents with unrestricted access. Customers may not give their password or one-time codes to such agents, which in principle may not initiate payments. Citizens warned customers by email that they too may be liable. JPMorgan Chase and Bank of America already announced similar policies in October.

US consumer rules that require banks to investigate unauthorised transfers cannot be waived by contract, Decrypt notes, but it is unclear how they apply to actions by an agent the customer has enabled. On 22 October, major banks reportedly saw AI agents logging into customer accounts and sometimes doing more than asked. TechCrunch also points out that banks see such tools as competition, because they can help customers avoid fees or find better interest rates.

Spotted a mistake in this article? Let us know.