Banks can't tell which AI agents to trust. That's fixable.

Every time an AI agent acts on someone's behalf, a business has to decide whether to trust it. That decision is getting harder, and banks are on the frontline.

On October 22, Axios reported that major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts. Some of those agents appear to be acting on their own, without customer's instructions, and some banks have tracked agents contacting their call centers. Morgan Stanley, Bank of America, JPMorgan, Wells Fargo, and others are racing to contain the risk. One bank official told Axios the industry has grave concerns about agents getting into customer accounts and taking steps that could affect people's finances.

So, what's keeping banks from trusting agents, and what solutions exist?

Why banks can't tell who is acting

Most of these agents log in with the customer's own username and password, often with the customer's permission. To the bank, the agent looks exactly like the customer.

Picture two agents logging into an account one morning. One is flagging an unusual transaction in the bank's records and the other is making a separate purchase. At login, they look identical. The bank can't tell which agent it's dealing with or who built and runs it. It can't see whether the customer authorized the actions or what the agent is allowed to do, and the limitations. One executive told Axios they struggle to tell agents apart, let alone tell whether an agent was authorized by a customer, or if it's gone rogue.

When something goes wrong, there's no record of who acted, on whose authority, or why. That leaves the bank left reconstructuring intent after the action was committed, making it harder to keep their customers, and their transactions, safe.

So right now, banks are doing the only thing they feel like they can: shutting agents out or limiting what they can do.

Morgan Stanley has started blocking third-party agents. JPMorgan, Bank of America and Citi are restricting them, and most banks are moving toward read-only access.

The decision contains the risk, but it doesn't solve the problem, and blocking only works on agents a bank can detect.

The industry is naming a need that isn't being met, and banks know it. Several banks and industry executives told Axios they recognize the need for middle ground because customers will want to use these agents. For now, however, being too permissive is too risky.

Banks have every right to be cautious, and they should be. What the industry needs isn't a wave of new agents, it's a way for businesses to trust agents without giving up control or protections.

The missing layer is agent verification

Banks know how to trust people and businesses with "Know Your Customer" (KYC) and "Know Your Business" (KYB). Agents need the same: "Know Your Agent" or KYA.

Right now, banks only see a login. With agent verification, it sees a verified agent, tied to a verified person, carrying permissions that are checked every time it acts. That turns agent access from an all-or-nothing request to something banks can control safely and securely, according to its own risk tolerance and policies.

None of this requires the customer to hand over a password.

How Beltic does this

Beltic was built for agents from day one, and it meets the verification standards regulated institutions already work to.

Verification starts with identity on both ends of the connection. Every agent gets a verified identity tied to the platform that runs it, so the bank knows exactly what is connected. We then verify the customer behind the agent with compliance-grade identity checks, bind them to the agent, and keep proof that they delegated access.

Knowing who is acting isn't enough, though.

The more difficult question is whether this action, by this agent, for this customer, is allowed under the bank's rules. So before an agent acts, Beltic reads what it's trying to do and why. Every agent carried a trust score based on its operator, its track record, and its behavior across businesses.

Each transaction is checked against what the customer authorized and the bank's own policies before money moves. Every allow or block decision leaves an audit record for risk teams and regulators.

The result is that unapproved actions are stopped before they happen, and verified agents are welcomed instead of blocked.

Beltic is also neutral by design. We work across agent platforms and payment networks, so a bank doesn't need a separate integration for every agent its customers bring.

Banks don't have to choose between blocking agents and trusting them blindly. By verifying the agent, the person behind it, and every action in real time, they can say yes to the agents their customers want, and no to anything else.

That's what safely accepting agents can look like.