When the Agent Trades, the Customer Pays

The clause is short and the consequence is not. Citigroup updated its online user agreement this week to state that customers who give a third-party AI financial tool access to their accounts bear the losses from transfers and trades that follow, including losses that arise when the tool leaks credentials or is itself breached.

The change puts the risk of automation on the person who switched it on. It also puts the bank in the position of processing instructions it did not originate, cannot inspect and did not authorize, while disclaiming responsibility for how those instructions were formed.

Citi is not first. Bank of America and JPMorgan Chase made comparable adjustments last month, PNC followed on Nov 18, and Citizens emailed customers this week to tell them the terms had changed. Four of the largest retail institutions moving within weeks of one another is a pattern rather than a coincidence.

The commercial logic is easy to follow. Banks have spent a decade opening account data to third parties under pressure from regulators who wanted customers to be able to move their financial lives elsewhere. The same interfaces that let a budgeting app read a balance also let an agent initiate a payment, and the liability question for that second use was never settled.

An AI agent is a different kind of counterparty from a budgeting app. A budgeting app reads; an agent acts, often through a model that decides what to do next rather than through a rule the customer wrote. When the instruction that produces a transfer is generated at runtime by a system nobody at the bank can audit, the ordinary chain of authorization gets thin.

The clauses also arrive as banks are deploying agents of their own. Several large institutions have announced assistants that summarize accounts, flag suspicious charges and in some cases move money between a customer’s own accounts. A bank that offers an agent and disclaims agents is drawing a line at the boundary of its own product, which is defensible and also convenient.

Consumer-protection rules limit how far a contract can reach. Electronic-fund-transfer rules in the United States require institutions to investigate errors and to restore funds in defined circumstances, and a term in a user agreement does not override a statutory duty. Where a customer was tricked into granting access, the analysis usually turns on whether the bank’s systems had reason to flag the transaction, not on what the agreement says.

That distinction is the one customers are likely to test. A tool that leaks credentials is a security failure on someone’s part, and the question of whose depends on how the credentials were stored and whether the bank’s authentication was bypassed. Contracts allocate cost between sophisticated parties; they do not decide facts.

For the companies building these tools, the practical effect is a shift in the insurance market. A vendor whose product can lose a customer’s money now needs to show that it carries coverage, because the customer’s bank has just told the customer that the loss is theirs to recover. That requirement travels up the chain to whoever hosts the model.

There is a version of this that regulators may prefer. Open-banking regimes in Britain and Europe have spent years building a framework in which third parties are registered, audited and liable for the instructions they send. The American model has relied more on contract, and these four banks are now writing the contract that the framework would otherwise have supplied.

The timing relative to adoption is what makes the change notable. Agents that trade are not yet common, and the banks are setting terms before the disputes exist rather than after. That is unusual, and it suggests the institutions expect volume quickly enough that they would rather be early than fair.

What the customer experiences is likely to be small and cumulative. A new disclosure at the point where a third-party connection is authorized, a checkbox, an updated agreement in an inbox. The practical question is whether anyone reads it, and whether an institution that has just disclaimed liability will still act quickly when a customer reports that an agent emptied an account.

The answer will come from the first serious case. If a bank reverses a transfer made by an agent that was given legitimate access, the clause is decoration. If it declines and the customer sues, the terms get tested against the statutory duties they purport to displace, and the outcome will set the template for the rest of the industry.

For now the four institutions have done the same thing and said very little about it. None of them has published the reasoning, and the emailed notice to Citizens customers described a change in terms rather than a change in policy. The customer who wants to know what happens if the agent errs will find the answer in a document they agreed to without reading, and will discover the rest in the first dispute.

Related Posts

  • November 20, 2026
  • 13 views
Sandisk Completes Eight-for-One Split as NAND Outlook Darkens

Eight for one. Sandisk’s stock split took effect on Nov. 19, and the shares began trading on the adjusted basis on Nov. 20, a change that divides the price by…

  • November 20, 2026
  • 13 views
A Lawmaker’s Privacy Suit Puts Model Design on Trial

For three years British courts have sidestepped the question of whether the company that builds an AI model answers for what its users produce with it. A written defence filed…