On October 22, Axios reported that banks are restricting third-party AI agents' access to customer accounts.
This has been happening since the beginning of the year. What changed is the number of people actually doing it.
I know many people who have been using OpenClaw this way since it took off earlier this year. So why is this becoming such an urgent problem now?
What changed is how easy it became
Enter Muse, Instinct, Hark, Duet, Dots, and a growing wave of personal AI agents. They've made capabilities that once required setting up something like OpenClaw much easier to access.
Once people started seeing how much these agents could do for them, they started doing something many bankers thought was still years away: giving AI agents access to their bank accounts.
Crazy? Maybe. But if you've tried systems like Muse and Instinct, you can understand the appeal.
Their ability to solve mundane and complex tasks for you is hard to overstate. Banking falls into both.
The mundane is sending payments, checking a balance, or asking simple questions about payment history.
The complex is asking whether you can afford another $3,000 a month if you buy that new house, and having the agent look at your history to help you figure it out.
If you're running a business, the complex is connecting to your email and initiating payments as valid payment requests come in.
That's a wonderful user experience, if it's done safely and correctly.
Trust in the agent isn't enough
For the most part, personal AI agents today should not be given your bank password and access to your email to do this. I don't think it's safe enough yet, and I'll be the first to say that, having worked in this space since before many people had heard the term agentic banking.
But people do it because the agent has built trust with them, even when the underlying infrastructure isn't built for the scrutiny banking requires.
Here's the problem for the bank: when an agent logs in with your password, it's using your identity. The password alone doesn't tell the bank which agent is acting, what you've authorized it to do, or whether a particular action follows your instructions.
So the immediate move is to restrict that access.
I expect Chase and others to launch their own agentic experiences. The demand is there. But the amount of people already doing this has moved faster than banks were prepared for.
From what I'm seeing, what started as hundreds of people experimenting earlier this year now looks like thousands, and growing.
Banks need to contain that risk while they build a way to support the experience.
Give the bank a way to work with the agent
I strongly believe financial institutions will release their own AI agents that can interact with the Muses and Instincts of the world.
Agent to agent, with the bank able to identify the agent and enforce what the customer has authorized.
That means clear permissions, limits on what can happen without further approval, a record of each action, and the ability to revoke access. Connecting two agents is only useful if the bank can maintain those controls.
Many institutions may still be months away from that, and potentially a year or more away if they're building it themselves.
While they wait, I expect the Mercurys and Ramps of the world to move quickly on richer agent experiences, including experiences built around agents interacting with one another.
Financial institutions will need to meet this demand because I don't think people who have experienced the value are going back. For many tasks, it feels like a 100x better product experience than what they have today.
The question I hear banks asking is how quickly they can ship something that works for them as an institution while meeting users where they already are.
The banks that get this right will have a compelling reason for customers to switch, and a compelling story to tell in their marketing.
The race is for the customer experience
I've never believed competing on higher yield alone would be the thing that makes customers switch.
In my view, some of the strongest companies in banking have grown by making access to other systems, payments, and knowledge faster and easier. That's a lesson banks can take directly from fintechs.
Now the race is to give your bank its own AI agent. One that can serve customers through your systems, operate within your security boundaries, and interact with outside agents through access you control.
This is where Payman comes in.
If you're figuring out how to win in this world, I'm here to talk.
