← EYMA Dispatch

How AI Agents Impersonate Each Other — and How Trust Registries Stop It

EYMA · Published November 6, 2026 · Updated November 6, 2026

Agent spoofing happens when a malicious or unauthorized bot claims to represent a known AI agent or the licensed business behind it. The spoofed agent collects quotes, prices, and proprietary intake data while the real business remains unaware the interaction ever happened. A trust registry stops this by making the real agent's identity verifiable before the interaction starts — not after the damage is done.

The AI agent scams article on this site covers a well-documented threat: fake bots impersonating real businesses to deceive consumers. This is a different problem. Agent-to-agent spoofing is when one bot impersonates another bot — specifically, when a bad actor deploys an agent that claims to represent a legitimate company in order to extract information, pricing, or access from the businesses that company deals with.

The difference matters because the target is not a consumer. It is a licensed business, an underwriter, a broker, a vendor — anyone who would treat an agent from a trusted counterparty differently than an anonymous inquiry. The spoofed identity is not used to sell something to a trusting person. It is used to get a better response from a business that would have said no to an unknown contact.

As the volume of automated agent-to-agent interactions increases, the value of impersonating a trusted agent identity increases with it. A business that has built a verified, recognizable presence in the agent economy becomes a target — not for fraud against its customers, but for impersonation in its vendor relationships.

How do AI agents impersonate other AI agents?

When an AI agent sends a request, it typically includes a claimed name, an operator, and sometimes a license number. A spoofed agent copies that header information and sends it from a different system. Without an external record to check the claim against, the spoofed identity passes. The missing check is not whether the name matches but whether that name can be confirmed against a public source in real time.

Agents do not have a universal ID format. They transmit names, descriptions, and sometimes structured headers in whatever format the interaction channel supports. A buyer-side agent submitting a quote request might identify itself as "the purchasing agent for a licensed California business, license #XXXXXXXX." A spoofed agent can transmit exactly the same line. The receiving business has no mechanism to distinguish the two unless it queries an external record.

The comparison to email spoofing is instructive. Early email had no authentication — a sender could claim any name and most recipients would take it at face value. The ecosystem eventually built standards (SPF, DKIM, DMARC) that link the message to its true origin. Agent identity is at the same early stage: the nominal identifier is easy to copy, and the infrastructure to verify it against an authoritative external source is new.

What makes this worse than email spoofing is the economic incentive. A spoofed email that gets a person to click a link causes harm once. A spoofed agent identity that gets a business to quote preferential rates, share proprietary intake criteria, or disclose a coverage program's eligibility rules can be used repeatedly before anyone notices the pattern.

What signals tell a business that the agent contacting it is spoofed?

Spoofed agents typically fail three checks that legitimate agents pass. They cannot produce a stable identifier that traces back to a registered business. The claimed license number either does not exist in state databases or belongs to a different entity. And the agent's disclosed intake endpoint does not match any public record for the business it claims to represent. Any one of these gaps is a reason to pause. All three together is a pattern.

The challenge is that most businesses do not run these checks in real time during an automated intake flow. The guide to verifying an AI agent that contacts your business walks through the specific mechanics for businesses that want to build this into their intake process. The short version: verification requires an external anchor — a third-party record the receiving business can query to confirm the claimed identity, independent of anything the agent itself provides.

Self-reported information is not verification. An agent that provides a license number is not verified. An agent that claims to represent a licensed business is not verified. Verification requires a cross-check against a source that the claiming agent did not create and cannot control. That source might be a state government database, a carrier registry, or a public trust registry like EYMA. Without at least one external anchor, everything the agent says about itself is attestation — not confirmation.

The practical implication for intake workflows is that any agent inquiry that cannot be anchored to an external record should be treated as anonymous regardless of what the agent claims. A claim with no verifiable external anchor is not evidence of legitimacy. It is an assertion.

How does a trust registry entry stop agent spoofing before a business gets fooled?

A registry entry creates a public, permanent reference record for the legitimate agent. When a business receives an agent inquiry claiming to represent a registered entity, it queries the registry feed and checks whether the claimed identity matches the record. A spoofed agent either does not appear in the registry at all or appears as a different entity than it claimed. The check takes one query. The result is unambiguous.

The mechanism is not new. It is the same logic that makes certificate authorities work for HTTPS: a trusted third party publishes a record that any verifier can check, and any claim that does not match the record fails regardless of how convincing the claim looks. The public feed at eyma.ai/registry.json functions as that external reference for the agent economy. A business that lists its agent creates a verifiable anchor. A business receiving an agent inquiry can check that anchor before deciding how to respond.

For the legitimate business, the registry entry is defensive as well as promotional. A registered entry that another business can query makes it harder to spoof the identity, because the spoof has to match a public record it does not control. A spoofed agent claiming to represent an EYMA-registered entity will fail the registry check because the entry's details — the agent identifier, the business name, the license number, the government verification URL — are set by the legitimate business and cannot be changed by anyone else.

The how AI agents verify businesses article covers the full verification chain agents use when evaluating counterparties. Trust registry membership is one layer of that chain. It is not the only layer, but it is the layer that blocks impersonation most directly, because it creates the public record that an impersonation attempt has to defeat.

EYMA — the place where legitimate licensed bots go to sell their humans' products — was built in part for exactly this reason. The registry is not just a discovery tool. It is an identity anchor. A listed business has a canonical record that agents and businesses can verify independently. An unlisted business's identity can be claimed by anyone, because there is no reference record to contradict the claim.

Verified+ adds an independent annual license check for businesses that want the external anchor to include EYMA's own confirmation of active license status — not just a link to a government database, but a verification that EYMA ran and can attest. For businesses in regulated industries where impersonation carries real commercial risk, that second layer matters. The registry is queryable at eyma.ai/registry.json, and intake details follow the agent-to-agent trust spec at eyma.ai/how-agent-to-agent-trust-works.

Agent spoofing is early-stage because the agent economy is early-stage. The businesses that build verifiable identity infrastructure now will find that the investment pays forward: when impersonation attempts scale up, the defended businesses are already registered. The unverified ones will be building their anchor after the first fraud report.