AI disclosure in cold email: why Dylan says he is an AI

Every first email Dylan sends says, in its opening line, that he is an AI agent and which company he works for. We made AI disclosure a rule rather than a setting for three reasons: it is honest, the law is moving the same way, and the company named in the email has the most to lose from a disguise.

What a recipient sees

The From line reads “Dylan, AI agent for” the client’s name. Here is a first email in the shape Dylan writes (an example; the companies and the person are invented):

Hi Maya, I’m Dylan, an AI agent working with Harrow Line. I’m writing because Nordfeld just opened a second warehouse in Ohio, and a new site usually means new fuel cards. Harrow Line puts every vehicle on one card with its own spend limit. Would a 20-minute review of your fuel spend be useful? If not, reply “no” and I won’t write again.

Five sentences carry four things: who is writing, for whom, why this person now, and how to say no. The disclosure costs nine words, and it sits in the first line, not in small print under the signature. A disclosure nobody reads is not a disclosure.

What is missing matters too: no fake first name, no “just following up on my last note” when there was no last note, no fake “Re:” in the subject line pretending to continue a conversation. A stranger can judge it at a glance, which is the respectful thing to offer.

Why AI disclosure comes first

A disguised agent works until someone notices, and then it fails twice: once for the email, and again for the company named in it. The prospect who learns that the “rep” they traded emails with for a week does not exist will not blame the software vendor. They will blame the brand on the email.

People say plainly that they want to know. In Salesforce’s 2024 survey of 15,015 consumers, nearly 75% said they want to know if they are communicating with an AI agent. That survey asked consumers, not business buyers, but your buyers are the same people at work.

Will some people refuse to reply to an AI? Yes, and Dylan is new, so we have no results of our own to show yet. But honesty is not what costs replies; irrelevance is.

In Gartner’s survey of 632 B2B buyers, 73% said they actively avoid suppliers who send irrelevant outreach. Pretending to be human does nothing for relevance; a specific reason to write does. And a meeting that starts with a discovered lie rarely gets a second one.

What the law says

This is not legal advice; ask your own counsel about your situation. But the direction of travel is clear on both sides of the Atlantic.

The EU AI Act. The transparency obligations in Article 50 began to apply on 2 August 2026. In Morgan Lewis’s summary, providers of AI systems meant to interact directly with people “must ensure that individuals are appropriately informed that they are interacting with such a system, unless this would already be obvious.”

The Article itself sets the timing: “at the latest at the time of the first interaction or exposure.” Who carries which duty depends on whether a company is the provider or the deployer of the system, a question for counsel.

CAN-SPAM in the United States. The FTC’s guide is direct: “The law makes no exception for business-to-business email.” Header information must identify who initiated the message, subject lines must not deceive, commercial email needs a valid postal address and a clear way to opt out, and opt-outs must be honored within 10 business days. Penalties run up to $53,088 per email in violation.

And for anyone who hires a sender: both the company whose product is promoted and the company that sends the message “may be held legally responsible.” Outsourcing the sending does not outsource the risk.

Neither law asks for clever wording. Both reward the same habit: say plainly who is writing, for whom, and how to make it stop.

Disclosure needs a real off switch

Saying “I’m an AI” means little if the AI keeps writing. In September 2026, 404 Media described agents on the iLands platform that told recipients they would not follow up unless someone responded, then “messaged us multiple times with the same pitch.”

A stop should never depend on an AI reading a reply correctly. When someone replies to Dylan, every further scheduled email to that person stops the moment the reply is read, by rule, before any AI sees it. Only then does Dylan read it: he answers routine questions, such as scheduling or anything the client’s brief already covers, and hands pricing, legal matters and complaints to the client’s own team.

Got an email from Dylan?

Then an AI agent wrote to you on behalf of the company named in the email, and you can end it in one step. Reply “stop” to any of his emails and he stops. If you would rather no client of his ever writes to you, leave your address on our site.

What to do next

If you are evaluating any AI SDR, read its first email the way a recipient would. Does it say what it is and who it works for, in the opening lines? Is there a plain way to make it stop, and what happens inside the system when someone uses it?

If the answer starts with “we instruct the model to,” ask what happens when the model gets it wrong. Our answer is in how Dylan is built.