Banks and Bots at the Barricades

This newsletter is part of The Assetizer, GenTwo's thought leadership platform.


 

Last week I wrote about AI agents coming into wealth management, mostly as tools for advisors, but also as tools for clients, who can now use them to trade and move money on their own. Since then the largest banks in the US, and this week the UK, have begun blocking agents that log in with a customer’s password and try to move funds.

What makes this interesting is that it is the first time the industry has had to decide, in practice rather than in a white paper, whether a piece of software can be a customer. The answer so far is yes, but only on the bank’s terms. So this week I thought I would look at what the banks are doing and why, and ask what it means for the rest of us.

The Bouncers

In mid-October the big US banks started turning AI agents away from customer accounts. Schwab’s CEO told Bloomberg his firm was blocking "tens and twenties" of attempts a week, some of them from Meta’s Muse, which by then had 10 million users. Jamie Dimon said JPMorgan was doing the same, since "you don’t want rogue agents and there could be thousands of them." Citi, Bank of America and Morgan Stanley followed. Last weekend the Telegraph reported that the big UK high street banks had joined in.

How did the bots get in? Mostly because customers gave them the password. Pay my bills, move my cash, find me a better rate.

And why did the banks turn around and throw them out? Generally for the obvious reason: a bank cannot tell whether it is you logging in or a program with your password, and all its security rests on that being you.

Dig deeper though and you will find two underlying concerns.

There is of course the worry of fraud, something which PNC’s chief executive says he is "scared to death" of. It is the bank’s duty to protect its customers against theft of their funds.

But those banks are also afraid for their own money. The Financial Times reported that what really worries the banks is agents sweeping idle cash out of accounts into whatever pays more. This is exactly one of the potential hidden money traps of AI agents I talked about last week.

No Doesn’t Always Mean No

On the other hand, nobody is pretending the agents will go away. That has led to two different kinds of response. Some banks have built their own client-side agents. Most still allow read-only access through providers like Plaid, so a bot can look but not touch. Schwab has launched a house assistant called Charley that answers questions but cannot trade. Bank of America’s Brian Moynihan wants a human to press the button before any money moves.

Wells Fargo, however, has chosen a different route. From 24 November its terms say that customers are "solely responsible" for any agent they use, and that whatever actions the agent carries out, "including those you did not want or intend," count as authorized.

Stay In The Loop

Last week I argued that regulators and clients would keep people in the loop for a long time yet. The banks have now shown the same instinct, though for mixed reasons.

For now, that’s probably just as well. Visa recently tested 22 AI shopping agents and found they got the job right only 57 percent of the time, and some of them made a purchase after reporting they had failed.

For wealth managers, this is evidence that the job is safer than the headlines suggest. Someone has to be answerable before money moves, and the banks have just made clear they do not want to be that someone for an outside agent. That is the advisor’s job. If anything they make it worth more, because the better the bots get at proposing, the more it matters who is allowed to approve.

That is a good thing. The agents will get better. The people they work for should still have to say yes.

Tom Lyons, GenTwo