Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Citi, PNC and Chase Rewrite Terms So AI Agent Losses May Fall on Customers

Citi, PNC and JPMorgan Chase have updated their account terms to set out who bears the loss when a third-party AI agent touches a customer's money. The terms suggest transfers can still count as authorized even when an agent ignores instructions or when leaked credentials are used by someone else.

Published
花旗、PNC与摩根大通改写条款:AI代理转错账,损失可能算客户的
Image source: citibank.com.sg

Citi, PNC and JPMorgan Chase have each updated their account terms to spell out who pays when a third-party AI agent makes a mistake. None of the three announced a new AI product; instead they used contract language to answer a very practical question: if an agent moves money on a customer's behalf, who owns the loss?

Citi's position is the most pointed. According to the terms, if a customer grants an AI agent access to an account or permission to carry out instructions, a resulting transfer may be treated as authorized by the customer even when the agent did not follow those instructions. The language also covers compromised credentials: if a password leaked through an AI agent is then used to log in or move funds, Citi may still treat the transaction as the customer's.

PNC takes a narrower route and tightens the doorway. It requires customers to use bank-approved connections for outside tools, and its terms generally bar external agents from paying bills or transferring money. Being able to see an account, in other words, does not mean being allowed to move funds out of it.

JPMorgan Chase draws its line at credentials. It permits customers to share account information through approved connections, but prohibits handing banking usernames and passwords directly to external AI tools. Data may flow; authentication secrets may not.

All three share one idea: a transaction an AI agent botched is not the same thing as a transaction the customer never authorized. Unauthorized transactions usually still travel through a bank's fraud and recovery machinery. Agent errors, under these terms, are pushed toward the customer, on the argument that the customer chose to hand account access to software. In practice, that shifts the cost of an agent's mistake from the bank's fraud team to the customer.

The timing matters because agents are only now reaching personal finance. Over the past year, tools that compare prices, book travel, manage subscriptions or initiate payments have multiplied, and most of them depend on users surrendering account credentials to a third party. Banks tightening their terms effectively adds a compliance cost to scaling that model.

For consumers, the practical takeaway is unglamorous: before connecting an account to any agent, read how that bank defines authorization, what it says about leaked credentials, and which actions are banned outright. For the industry, the open question is whether a patchwork of bank-specific clauses converges into a common standard, and whether consumer-protection regulators decide to weigh in.

Why it matters

By assigning agent-caused losses to customers, banks make third-party finance agents riskier for consumers and more expensive for the startups building them. If other institutions copy these terms, liability language becomes the real gatekeeper for AI agents in personal banking.

CitiJPMorgan ChaseAI AgentBanking
Back to realtime news

Nearby Updates

All