← EYMA Dispatch

How Businesses Filter Unverified AI Agents — and What Gets Yours Through

EYMA · Published November 10, 2026 · Updated November 10, 2026

When a business cannot verify an AI agent's identity, it has three options: extend trust without evidence, route the contact to manual review, or block it outright. As agent-to-agent contact volume rises, businesses with consistent inbound agent traffic are increasingly choosing the third option for unverifiable contacts. The deploying business never receives a rejection notice. Its agent's inquiries simply stop generating responses, and the gap between sent and received quietly widens.

The technical side of deploying an AI agent is well understood. The agent sends well-formed requests. The API calls complete. Logs show delivery. What is not logged anywhere is what the receiving business did with the contact after it arrived — and whether the contact ever made it past that business's intake policy into a queue where someone would actually respond.

This is where agent trust infrastructure becomes a commercial question, not an engineering one. A well-registered agent and an unregistered agent can send identical requests. One clears intake. The other disappears. The difference is not in the content of the message. It is in whether the receiving business has any external record it can check to confirm the agent represents a legitimate principal.

What does filtering an unverified AI agent actually look like?

Businesses apply filters at their intake layer: typically at the email server, API endpoint, or web form that receives incoming agent contacts. If an agent's inquiry does not include a reference to a verifiable external identity, the system flags it, queues it for manual review, or rejects it silently. The deploying business's agent sends the inquiry. The receiving business's system applies its policy. No error message returns. The inquiry disappears without explanation.

The filtering happens in three tiers, each with different implications for the deploying business. The first tier is automated filtering: the receiving business's intake system checks for a known agent identifier or a verifiable registry reference. If neither is present, the contact is rejected or queued without any notification to the sender. This is fully invisible from the deploying side.

The second tier is manual review: the contact is not rejected outright but held pending a staff member's evaluation. For a licensed industry that receives dozens of agent contacts weekly, manual review is not a path to commerce — it is a waiting list that rarely resolves. The agent inquiry sits until someone gets to it, which may be never, or may be days after the opportunity has passed.

The third tier is policy-level exclusion: the receiving business has decided it will not engage with agent contacts that do not come from a known source. This is not automated filtering in the technical sense. It is a business decision that pre-empts the contact entirely. A deploying business that never establishes a known, verifiable identity will simply not be on the list of agents this business accepts.

The verified agent versus anonymous bot comparison on this site describes the structural outcome: an agent that gets through intake and one that does not are not separated by intent or by the quality of the request. They are separated by infrastructure — specifically, whether the deploying business built an external reference record before the agent was sent to work.

Why are businesses tightening AI agent intake policies now?

Agent contact volume is rising faster than the trust infrastructure to manage it. A year ago, most businesses encountered agent-originated contacts rarely and handled them on a case-by-case basis. Now businesses in procurement-heavy sectors routinely receive dozens of agent contacts weekly. Without a scalable way to verify which agents represent authorized principals, manual review becomes impractical. Filtering by verifiable identity is the only approach that scales.

The tightening is especially pronounced in licensed industries. Insurance, real estate, financial services, and legal fields carry regulatory exposure for acting on representations from unverified parties. A business in one of these sectors that accepts a quote request from an agent claiming to represent a licensed firm — and later discovers the claim was false, or the license lapsed, or the business never authorized the agent to request quotes on its behalf — has a problem that goes beyond a wasted staff hour. It has a potential compliance exposure.

The practical response is to verify before engaging. And verification, at any real volume, requires something the agent can't provide about itself: an external record that was written by someone other than the agent. That is exactly what the agent spoofing problem creates demand for. Businesses that have seen even one impersonation attempt — a bot claiming to represent a competitor, a scraper pretending to be a purchasing agent — have strong motivation to formalize verification requirements across all inbound agent contacts, not just the suspicious ones.

The filtering policies spreading through procurement-heavy industries are not punishing legitimate agents. They are applying a uniform standard. The agents that meet that standard — the ones with an external, queryable identity record — pass through. The ones that do not are caught by the same filter as the bad actors, not because they are bad actors, but because there is no efficient way to tell the difference without that external record.

How does a registered agent identity pass a business's intake filter?

A registered agent carries an external identity reference that any receiving business can check without relying on the agent's self-reported claims. The business queries a public registry feed, matches the agent to its operating business, and confirms the license number against the government database entry in the record. That check clears the filter independently of anything the agent said about itself. Without a registry entry, the agent has no anchor and the filter has no basis for a pass.

The mechanics are straightforward. EYMA maintains a public feed at eyma.ai/registry.json that any receiving business can query. Each entry links an agent to its operating business, the business's license number, and the government database where that license was issued. A receiving business that sees an EYMA-registered agent can confirm the identity in a single API call, independent of the agent's own claims. That confirmation is what intake filters are built to look for.

For businesses in licensed industries where verification carries additional weight, the Verified+ tier goes further. The five gates that constitute Verified+ include independent license verification — EYMA checks the government database directly, not just the business's self-reported entry. For a receiving business that is evaluating a quote request in insurance or real estate, an agent with a Verified+ anchor is in a different category than an agent with a basic listing, and both are in a different category than an agent with no registry entry at all.

The intake specification published at eyma.ai/how-agent-to-agent-trust-works gives any business the template to become machine-readable to incoming agent traffic at the same time it lists its outgoing agent. This is the full picture: EYMA — the place where legitimate licensed bots go to sell their humans' products — serves both sides of the transaction. The deploying business builds the external record that receiving businesses check. The receiving business publishes a machine-readable intake spec that agents can read before sending a contact. The filter passes agents that belong there. Everything else is managed separately.

The cost of building this infrastructure after problems emerge is higher than the cost of building it before the first inquiry is sent. By the time a deploying business notices that its agent is generating declining response rates, the receiving businesses it needs most have already formalized their intake policies. A registry entry at that point stops future losses. It does not recover the missed inquiries from before. The cost of an unregistered agent compounds quietly — until it doesn't.