Citi has updated its Online User Agreement to make customers responsible for activity conducted through third-party AI agents they authorize, according to TechCrunch’s November 18, 2026 report. The reported change goes beyond an assistant paying the wrong bill. It also addresses losses when credentials entrusted to an AI tool are subsequently leaked or compromised.
JPMorgan, Bank of America, PNC and Citizens have issued similar terms or warnings, TechCrunch reports. The practical issue is immediate for anyone considering connecting a financial assistant to a bank account: granting access may carry consequences that outlast the original instruction.
Citi also distinguishes outside agents from tools it provides itself. That makes the provider, access method and permissions important parts of the decision, alongside whatever the AI promises to do.
What Citi’s Updated Terms Cover
Citi’s spokesperson told TechCrunch that the agreement reinforces that:
“customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents.”
Citi’s public Online User Agreement governs electronic services accessed through Citi Online and Citi Mobile. It also says the underlying accounts and services are governed by separate agreements. That layering matters: an AI-related provision in the online agreement may sit alongside account-specific rules and transfer terms.
A published breakdown by Unite.AI describes the updated agreement’s Section H as applying broadly to third-party software taking autonomous, semi-autonomous or non-autonomous actions for a customer. On that account, the provision is not limited to a fully independent agent that manages someone’s finances without supervision.
The detailed Section H and funds-transfer provisions discussed here come from that secondary breakdown; TechCrunch separately reports the central liability shift and carries Citi’s response.
According to Unite.AI, supplying credentials or otherwise permitting an agent to access an account causes Citi to treat the agent’s access and actions as customer-authorized. The summary says customers must monitor those actions for accuracy, completeness and compliance with the authority they intended to grant.
The important distinction is between what you asked the agent to do and what access you gave it. Citi’s reported terms give account access consequences that a narrow prompt may not undo.
