Opinion: Banks Are Right to Want a Human Sign-Off on AI Agents. They Are Wrong to Lock the Door.

LOS ANGELES - Start with the part the banks get right. An AI agent that can log in with your password can do anything you can do, at machine speed, on a bad day for the model. Brian Moynihan’s line this week, that there should be “a human in the loop” before money leaves an account, is not a dinosaur talking. It is the same principle the Financial Stability Board wrote into its guidance on Thursday, and it is the same principle anyone who has watched a chatbot confidently book the wrong flight would endorse.

Jamie Dimon was also right that handing an agent your credentials is a “liability transfer.” Under today’s rules, if a bot you invited in drains your checking account by mistake, you may struggle to argue that the bank should make you whole. Banks did not invent that problem, but they are the ones who will hear about it first, and loudest. Wanting a clear answer on who pays when an agent misfires is reasonable.

Photo: InSapphoWeTrust / Wikimedia Commons (CC BY-SA 2.0)

Now the part the banks would rather you skip. The restrictions arrived in the same few weeks that Apollo’s Torsten Slok asked whether an “agentic bank run” was coming, and that question was about money, not malware. His argument was simple: an assistant like Meta’s Muse could notice your checking account pays about 0.1 percent, find an account paying 3.3 to 5 percent and move the cash for you. JPMorgan, Bank of America and Wells Fargo sit on roughly $1.6 trillion of deposits that pay nothing, the Financial Times has reported. At 3 percent, that is close to $48 billion a year in interest their customers do not earn. You do not have to be cynical to notice that the most dangerous thing an agent could do, from a bank’s point of view, is exactly what a good financial adviser would tell you to do.

Look at what the policies actually allow. At Citi and JPMorgan, an agent connected through an approved aggregator can read your balance. It cannot move a dollar. That is the safest possible design for the bank’s deposit base, and it happens to be the least useful design for the customer. An agent that can tell you your money is idle but cannot do anything about it is a notification, not an assistant.

Photo: Epicgenius / Wikimedia Commons (CC BY-SA 4.0)

There is a better answer, and banks already know how to build it. Customers should be able to grant an agent scoped, revocable permission: move up to a set amount, only between accounts I name, only with a push confirmation on my phone above a threshold, with a log I can read and a kill switch I control. That is how card networks treat recurring payments and how open-banking rules in Britain and Europe already treat payment initiation. It keeps the human sign-off that Moynihan wants where it matters, on new payees and large sums, without pretending that every transfer between my own accounts is an attack.

The industry has been circling a version of this idea. When Meta, Sierra, Shopify, Stripe and others proposed a common protocol for personal AI agents earlier this month, I argued that the rulebook should prove who the agent works for. Banking is where that proof would matter most. A verified agent, acting under limits the customer set, with liability rules written down in advance, is a product. A flat ban is a moat.

Photo: Suiren2022 / Wikimedia Commons (CC BY-SA 4.0)

The banks say they are building their own tools, and JPMorgan has promised a Smart Cash feature that will sweep client money into higher-yielding products automatically. Good. But “trust our agent, not yours” is the oldest move in platform economics, and customers should recognize it. Regulators who are quietly cheering the restrictions should ask a harder question: if a bank’s own software is allowed to move my money for me, why is software I chose held to a stricter standard than permission, limits and a signature? Human sign-off is the right rule. Locking the door is not.

Sources: Financial Times, “Citi becomes latest US bank to restrict AI agents from accessing accounts,” Oct. 16, 2026, https://www.ft.com/content/f49c8e39-f07f-4177-9948-aabe55e5ed91 (Citi and Bank of America restrictions, $1.6 trillion non-interest-bearing deposits, regulators); CNBC, Brian Moynihan interview, Oct. 14, 2026; CNBC and Fortune, JPMorgan third-quarter earnings call coverage, Oct. 13 and 15, 2026 (Dimon quotes, read-only aggregator access); Apollo Global Management, Torsten Slok, “Is an Agentic Bank Run Coming?,” September 2026; PYMNTS, “AI Agents Wake Up $1.6 Trillion in Sleepy Bank Deposits,” Oct. 8, 2026 (Smart Cash); Financial Stability Board, final report on AI sound practices, Oct. 15, 2026.

---

Marcus Reid, Opinion. Marcus Reid is a senior columnist writing on gadgets, platforms, and consumer tech. Previously edited product commentary for independent magazines; based in Los Angeles.