Knock, knock… who’s there? AI agents

schedule3 Min Read

Remember the thing we used to do as kids?

Knock, knock.

Who’s there?

Agent.

Access denied.

That’s seemingly where some US banks stand on personal AI agents today amid discussions around customer security and the liability question if something goes wrong.

But part of the concern may not just be about security. It could also be about economics. For example, an agent that can work around the clock comparing rates could meaningfully shift household cash into higher-yield accounts. With a prompt.

The ID at the door

On the security front, on 15 October, Proof announced production availability of its agent identity wallet and an open x401 protocol to bring verified legal identity and signed transaction authorisation to AI agents.

The idea is simple. An agent may have a valid login credential, but that does not establish that the person behind the agent authorised a particular action. This creates a liability concern for banks, lenders, and insurers. As AI agents proliferate, organisations need verification of who is behind the agent, what that person has authorised the agent to do, and how the transaction gets paid.

Consider the following example. An AI agent requests a $25,000 transfer from a consumer bank account. The bank issues an x401 challenge, asking for verified identity and a signed approval covering the amount, the recipient, and the source account. The customer reviews and approves the details on their own device. The bank verifies the signed instructions against its own existing requirements and executes the transfer.

In other words, the agent no longer has to say, “Trust me.” It comes with the receipts.

Here comes the barrier

But connecting the Proof wallet to an AI assistant does not mean access is automatically granted. Just because the customer authorises an AI agent to act on their behalf does not mean the bank will allow the request. Reading balances, recommending a (new) account, and moving money are different activities with different risk profiles, and banks are right to treat them differently.

If a bank’s objection is missing evidence of identity and consent (hence liability concerns), x401 provides a way to verify. And if that evidence is available but access is still denied, then the objection may be about something else.

With AI agents, a customer can potentially delegate the grunt work of comparing rates on a regular basis and deciding when moving money is worthwhile.

Faced with the potential of deposit flight, banks have three options:

A well-designed agent is one that knows how to optimise within the customer’s constraints, not simply chasing the highest yield. Other factors such as the customer’s preference (brand value, for example) and fees also come into play. But setting up barriers does not resolve the underlying problem: agents challenge the economics of customer inertia.

Friction can buy time. But it is not a retention strategy.

Rather than “How can banks stop agents from moving deposits?”, it’s time to be honest about what makes customers stay when money movement and management become truly frictionless.

The agent is now at the door with its ID.

Knock, knock.

Someone should answer.

About the Author