Agents need a vouch

Society is built on trust. I trust you. You trust me. We can trust each other.

The internet has never really had a good way to do that, especially when it comes to automated activity. We can see the results — platforms are either flooded with spam, ban all agents, or are in a never-ending battle to figure out who can be trusted and who needs to be banned. That may have worked a few years ago, when most automated activity was scraping and spam, but that’s too harsh for a world where people are using agents to trade, book their flights, and manage their money.

Automated requests passed human traffic for the first time ever in June, 57.5% automated vs. 42.5% human. But according to the Thales/Imperva 2026 Bad Bot Report, only 40% of those can be considered “bad bots.”

Platforms need nuance. They need to be able to know which agents to trust, and which are the bad bots.

How do you vouch for an agent?

The verification vouch

A slide widely attributed to a 1979 IBM internal training presciently stated, “A computer can never be held accountable, therefore a computer must never make a management decision.” We feel the same way. You cannot trust an agent who is not someone’s responsibility, because they cannot take responsibility on their own.

Agents make decisions, but every decision is based on the prompting and direction of a human somewhere. They never start from zero; agents only act on the information given to them.

That’s why you need something more nuanced than a global ban, or letting all bots in without a filter. You need accountability.

Agents should always be linked to an accountable human. It helps good actors keep track of their agents, and bad ones from shirking responsibility for the damage caused.

That’s why you need a vouch. Imagine you’re hosting an event — say, a dinner party — and your friends, who you already trust, want to invite someone. You give them permission, because you already know them and trust their judgement. If their invitee is the life of the party, making friends, and generally contributes to the good atmosphere, great!

But imagine instead that the invitee is a real mess. They start fights. They steal. They throw wine. You don’t just want to kick them out of the party; you might want to second guess the judgement of the person who vouched for them. You might even avoid inviting that person again.

We think verifying agents should work the same way. For Know Your Agent, we use Very’s palm biometrics to link agents to a real, verified account with a real, verified person behind it. Users do a quick scan of their palm, link the agent to their account, and move on. Platforms receive a quick attestation that can be verified in milliseconds. Very reverifies it again at key, sensitive moments — like before a trade that exceeds a spending limit, for example — by asking the human to complete a few hand motions. All data is processed on the user’s own smartphone, since all the platform needs is the attestation. No external hardware needed.

It’s really only once we can vouch for an agent that we can trust it. And it’s once we trust them that the agentic economy can really take off.

And that’s good for everyone — people and agents.

(All em dashes in this post were created by a human)