News take
Who pays when your agent spends money it shouldn't?
By Pat Villaceran · 14 October 2026
Picture a sweater in the wrong size. An agent ordered it for a shopper. The bank approved the payment. Now what? That small example sits near the top of a piece in American Banker by Penny Crosman, published on 13 October. The bank's approval, she writes, doesn't prove the agent bought what the shopper asked for. So who is on the hook?
A bank asks the hard question out loud
Jackie Reses runs Lead Bank in Kansas City. She doesn't pretend this is simple. Here is how she put it to American Banker: "Who is responsible? The human who misprogrammed? The model that enabled the agent to go outside the control? Is it the bank that let the transaction go through when it clearly was flagged to be incorrect?"
I love those questions. They're the right ones. And she doesn't stop at asking. Her bank has drawn up a "waterfall" of responsibility. It's Lead Bank's proposal. As American Banker reports it:
- "In the bank's proposal, a consumer would not be liable where an agent exceeded delegated authority."
- "The bank would be liable where it had control over the delegation of authority and failed to enforce it."
- "An AI provider would be liable where failure arose from a model defect, insecure tool use or inadequate safeguards."
- "The merchant would be liable if it knowingly induced an agent to exceed its authority."
Read it slowly. Every kind of error is assigned to a party. The customer, the bank, the model provider and the merchant each know where they stand before anything breaks.
Records make it real
The same piece covers Primitive. It's a startup founded by Derek White, a former chief digital officer at U.S. Bank. By White's own description, its standard includes traceability for every agent session. He lists what it tracks: "Is it operating within governance? What LLM is it hitting? What are the token costs associated with that?" He says the trace can be packaged up for a regulator or an auditor.
In our upcoming episode on agentic AI, I talk about a mix-up over a reminder at work. "There was no record. But now we have all the records." A record doesn't decide who pays. It does show what happened. That makes the list possible.
It's a bank, and it's a proposal
You're not a bank. Your agent might book meetings, answer customers or reorder stock. And this is Lead Bank's proposal. But the question doesn't change with the industry. Something will go wrong. Who answers for it?
In that same recording, I said: "For me, Nathan, innovation is always about accountability. You can always blame someone else. You can always blame the market…" When an agent gets something wrong, blaming someone else is very easy. Everyone in the chain has a reason why it wasn't them. A list written before launch takes that excuse away.
Make your list before launch
Take one agent you plan to switch on. Write down every kind of error you can imagine it making. It might spend too much. It might act without permission. It might misread a customer, or get tricked by a supplier.
Next to each one, write who answers for it. Is it the customer, your company, your model provider or a partner? Then write what you'll keep as a record, so you can tell afterwards which line applies. Share the list with the people who build and run the agent.
The button below puts you on the waitlist.
← All articles