Banks, card networks, and regulators are deciding who answers for AI that acts for customers, and companies that wait will inherit their rules.
The AI Transformation Brief
This week, the companies that handle money started writing down who answers for AI that acts for a customer. JPMorgan's CEO told customers who give AI assistants access to their accounts that the risk moves to them. Banking Dive American Express published a way to record what a shopper approved before AI buys anything. American Express Bank regulators in the US and Europe are asking whether banks can see, limit, and switch off their AI. CTV News The firms that define these limits first will set the terms for everyone else.
On the October 13 earnings call, JPMorgan Chase CEO Jamie Dimon said customers who give AI assistants access to their bank accounts should understand there is a "liability transfer." He said JPMorgan will not let just any outside AI assistant into the bank, calling it too risky for clients and for JPMorgan. CFO Jeremy Barnum put annual AI spending closer to $700 million, called that figure "a little stale," and said AI benefits are not yet visible in quarterly earnings. Its own cash-sorting tool, Smart Cash, is in a trial. Banking Dive
The largest US bank just told customers where its responsibility ends, and that line matters more than the warning. Once an outside AI assistant moves money, the bank wants the customer to own the mistake, while it builds its own assistant to keep the relationship. Every company that holds customer accounts will face the same choice: which outside AI tools to allow, on what terms, and who pays when one gets it wrong. Write that policy now.
Responsibility sits with whoever approved the AI's access, which today means the customer by default. Companies that leave this unstated will absorb the losses anyway, through complaints, refunds, and regulators asking why no one drew the line.
American Express on October 13 published Intent Intelligence, its recommended method for recording a customer's purchase instructions before an AI assistant completes a purchase. The customer sets limits such as the merchant and a maximum budget, confirms their identity, and approves them. The assistant then pays with a limited payment credential tied to those limits, never the customer's real card number. American Express Hannah Bozian, an Amex vice president, said that if a customer asks for white Nike sneakers and the assistant tries to buy Adidas, Amex could stop the sale. TechCrunch
This moves the customer's approval from the end of a dispute to the start of the purchase. Whoever holds the record of what a customer approved decides who pays when AI gets it wrong. Amex wants to be that record and wants other payment companies to adopt the same method. American Express Retailers and banks should decide this year whether to rely on another company's record of customer approval or keep their own. Every business that sells online should test its checkout with an AI shopper now.
Reuters reported on October 13 that banking regulators are stepping up scrutiny of AI, citing six people familiar with the matter. The Federal Reserve, the Office of the Comptroller of the Currency, and the Federal Deposit Insurance Corporation are asking banks what limits they place on AI, how its actions are approved, whether its decisions can be tracked, and how to shut it down. They are also asking whether banks use Chinese-developed AI. The European Central Bank gave lenders an end-of-October deadline for plans to counter AI-enabled cyberattacks. CTV News
These questions read like a checklist every company should be able to answer, in banking or anywhere else. Can you list every AI tool acting for you, show what it may touch, prove who approved each action, and stop it on command? Most companies cannot do all four today. Regulators are still gathering information, which leaves a short window to answer on your own terms. Operations leaders should run that checklist against their five most-used AI tools this quarter and close the gaps.
Kroger plans to cut about 3,000 corporate roles over the next three years as it brings AI into its daily work, CEO Greg Foran told staff in a memo seen by Reuters on October 12. The cuts would affect about 37% of its 8,200 corporate employees, with details coming in the months ahead. Foran wrote that Kroger should reduce duplicated work and that the savings would help cut prices and invest in store employees. Kroger cut about 1,000 corporate roles in June last year. KSL.com
The useful signal is where the savings go. Kroger is shrinking the office to pay for lower prices and better-staffed stores, which turns AI into a way to move money toward customers. The risk is cutting people before the new way of working is proven. Leaders planning similar cuts should first decide which tasks AI will take on, which stay with people, and who checks the results. Then judge the effort by prices, service, and store performance.
Palmer's "When Your AI Has Agency" (October 11) predicts that more purchases will begin with a standing instruction to an AI assistant and end with an approval request or a receipt. He writes, "Remembering something does not confer permission to act on it." Shelly Palmer That is the gap American Express and JPMorgan moved to close this week, from opposite sides of the purchase.
Decide where your responsibility ends when a customer's AI assistant acts for them.
Write the terms for outside AI access, name an executive owner, and publish the policy before a costly mistake forces the decision.
Power is moving to whoever keeps the record of what a customer approved.
Card networks, banks, and large retailers are each competing for that role. Companies without their own record will live by someone else's rules and fees.
Answer four questions for every AI tool: what it may touch, who approved its actions, whether its decisions are recorded, and how to stop it.
Close the gaps this quarter.
Ask management for a one-page list of every AI tool that can move money, sign agreements, or change customer records, with an owner and a way to stop each one.
Review it every quarter.
This week, responsibility for AI was assigned in writing: by a bank to its customers, by a card network to a purchase record, and by regulators to bank leaders. The assumption this breaks is that AI mistakes can be sorted out after the fact. The decision it forces is which approvals your company will record and own, and which it leaves to a bank, a card network, or a regulator.
When a customer's AI assistant makes a costly mistake with your company, can you prove who approved it?
Make AI accountable. Measure the results. Redesign the business.