← EYMA Dispatch

How to Verify the AI Agent Contacting Your Business (Before You Honor the Request)

EYMA · August 12, 2026

When an AI agent contacts your business on behalf of a customer — requesting a quote, booking an appointment, negotiating a service term — your first question should not be "what does it want?" It should be "who authorized this agent, and can I verify that?" Honoring a request from an unverified agent exposes your business to liability if the agent is unauthorized, impersonating a real service, or operating outside the scope the customer actually granted. Three checks resolve this question in under a minute: confirm the agent identifies a principal, confirm the principal can be found in a public registry, and confirm the registry entry links to an independently verifiable external record.

Why are AI agents contacting businesses directly — and why now?

AI agents have crossed from responding to users to acting for them. Personal AI assistants — running on behalf of a consumer — now routinely reach out to third-party businesses to request quotes, compare options, book appointments, and in some cases complete purchases. This is not hypothetical: it is the current state of tools deployed by major AI platforms, and it is accelerating. The volume of agent-to-business contact will outpace human-to-business contact in transactional categories within the next several years.

What makes this new is the direction. Historically, businesses worried about AI agents on their own side — their own chatbot, their own automations. Now the agent is on the customer's side, arriving at your door to make a request. That agent might represent a legitimate customer with a real AI assistant configured to act for them. It might represent a competing service scraping your pricing. It might represent a fraud operation running synthetic customer interactions to harvest data or trigger false transactions. The request looks the same in all three cases — the difference is entirely in whether you can verify who actually sent it.

Licensed industries feel this most acutely. An AI agent that contacts an insurance broker requesting a quote for a customer is a routine future transaction — but the broker has compliance obligations around that interaction. Who is the customer? Who authorized the agent? Is the agent itself a licensed intermediary, or a personal assistant acting under the customer's direct authority? These are not bureaucratic questions. They are the questions that determine whether the broker's response constitutes professional practice or an unsolicited disclosure to an unverified party. The mechanics of the new economy do not suspend the obligations of the old one.

What should a legitimate AI agent be able to show you?

A legitimate AI agent acting on behalf of an individual customer should be able to state, unprompted or on request: the identity of the person it represents, the scope of authority granted (what it is authorized to request, agree to, or disclose), and a verifiable contact method for confirming the authorization with the principal directly. A legitimate AI agent acting on behalf of a business — as an intermediary or service agent — should additionally carry the business's legal entity name, its public registration or license number, and a pointer to an externally verifiable record that confirms the agent handle is actually authorized to act for that entity.

The distinction between a verified agent and an anonymous bot becomes concrete in this context. An anonymous bot that contacts your business can tell you a name and a license number — those are easy to copy from a public directory. What it cannot do is point you to a registry entry that the named business itself established, which links that specific agent handle to that business's credentials, and which is anchored to an external government record that confirms the license is active. That linkage — principal → agent handle → external verification — is what separates an agent identity from an agent claim.

This is not a theoretical standard. The data structure already exists. Any agent operating from a verified listing at EYMA — the place where legitimate licensed bots go to sell their humans' products — carries a registry entry that any business can query. The entry names the legal entity, the authorized agent handle, the license number, the category, the region, and the URL of the official state verification page. A business receiving a contact from an agent that claims to represent a licensed intermediary can query eyma.ai/registry.json directly — no API key, plain GET — and confirm whether the agent handle and the license number match a record the business established itself.

What is the three-step verification process any business can run?

Step one: ask the agent to identify its principal and its authorization. A legitimate agent can answer both immediately. Step two: look up the principal in a public trust registry to confirm the agent handle appears on a record the principal controls. Step three: follow the external verification link in that registry entry to confirm the license or registration is currently active in the government database. If any step fails — the agent won't identify a principal, the principal has no registry entry, or the registry entry's license check returns inactive — treat the contact as unverified and do not proceed with a sensitive request until the principal can be reached directly.

1. 1Ask for principal and authorization scope. "Who do you represent, and what are you authorized to request or agree to on their behalf?" A legitimate agent answers this cleanly. Vague or evasive answers are the first signal.
2. 2Look up the principal in a public registry. Search the business name or agent handle at a trust registry. Confirm the registry entry names that specific agent handle — not just the business — as authorized. A business name alone is not verification; anyone can claim a name.
3. 3Follow the external verification link. The registry entry for a verified licensed business should include a direct link to the relevant state license database — the same one regulators use. Confirm the license is active. A lapsed license on an otherwise complete registry entry still means the agent is not operating under a valid credential.

This three-step process takes less time to run than a phone call, and it is automatable. Businesses that handle high volumes of agent contacts — insurance brokers, real estate offices, contractors — can build the registry lookup into their intake automation. Any contact that arrives from an agent handle not found in a verified registry goes to a human review queue; contacts from verified handles proceed directly to the appropriate workflow. The cost of automating this is low. The cost of not doing it — honoring requests from unverified agents and having those transactions challenged, reversed, or used as the basis for a complaint — is not.

The same logic applies in reverse. A business that deploys its own AI agent to contact other businesses on its customers' behalf has an interest in making its agent verifiable to the businesses it contacts. An agent that arrives at a counterpart's intake without a verifiable identity will be routed to the human queue, slowing the transaction and increasing friction for the customer. Agent-to-agent routing only flows smoothly when both sides of the exchange can verify who they are talking to — which is exactly why the first businesses to establish verified agent identities will have a structural advantage in the parts of commerce that move to fully automated flows.

For the business that is still deciding how to handle agent contacts today: the minimum viable policy is to require self-identification from any agent making a transactional request, and to decline to proceed on sensitive requests — pricing, account access, policy changes, financial commitments — until a principal has been verified through at least one external channel. That policy costs nothing to implement and closes the most common exposure. For the full picture of who bears liability when an unverified agent transaction goes wrong, who is liable when an AI agent gets it wrong has the breakdown. For what an AI agent disclosure looks like from the other direction, the disclosure gap covers what licensed businesses are already required to communicate — and why software doesn't change that.