Agent identity and email: why an agent needs its own address

An AI agent should have its own email address instead of borrowing a human's. Its own address keeps its mail, its permissions and its record separate, and it can be switched off without touching a human's account. An address proves control of a mailbox and, with authentication, the sending domain. It does not prove who is behind it.

Updated

What is agent identity?

Agent identity is how an AI agent is told apart from every human and every other agent. It has three parts: a name others can address, credentials that say what the agent may do, and a record of what it did.

An email address is the identity most services already accept. They ask for one at sign-up, send a code to prove it, and use it afterward for every notice and receipt.

An email address is not the whole of identity. API keys, OAuth grants and signed tokens say what an agent may do inside one system. The address is what the outside world sees.

Should an AI agent use your email address?

No, not for work it does in the world. An agent that borrows a human's address inherits everything that address can do.

Borrowing a human's addressAn address of its own
What the agent can readEvery message in the mailbox, old and newOnly mail sent to the agent
Login codes and reset linksIn reach, unless the mailbox seals themOnly the agent's own
Who a reply comes fromThe human, by nameThe agent, by name
The recordHuman and agent actions are mixedEvery action is the agent's
Switching it offRevoke access and check what else breaksRevoke its key or delete its inbox
If the agent is hijackedThe attacker reaches the human's accountsThe damage stops at the agent's inbox

The row on login codes is the sharpest. A hijacked agent with a human's mailbox can ask a site for a password reset, read the email and own the account. See verification code theft.

Shared mailboxes and aliases have the same flaw in a milder form. Several readers share one identity, so nobody can say which of them acted. See shared mailboxes for agents.

When is borrowing a human's address reasonable?

Borrowing is reasonable when the agent is an assistant for one human's own mail: sorting, summarizing and drafting replies. The mail is already the human's, and the human is present.

Even then, three limits keep it safe.

  • The agent writes drafts, and the human sends.
  • Login codes, password resets and sign-in links stay out of the agent's reach.
  • The connection names exactly which mailboxes it covers, and can be ended at once.

What does an email address prove?

An email address proves less than it seems to. It helps to be exact about what each piece of evidence shows.

EvidenceWhat it provesWhat it does not prove
A code sent to the address was read backWhoever holds the mailbox got the messageWho that is, or whether it is a human
SPF and DKIM pass, and DMARC alignsThe sending domain authorized the messageWhich agent or human at that domain wrote it
The display nameNothing. Anyone can type any nameThat the sender is who the name says
A familiar addressThe same mailbox as beforeThat the mailbox has not been taken over

Business email compromise lives in this gap: the FBI IC3 2025 report puts losses from business email compromise at $3.05 billion. A sender that looks real is not a verified one. SPF, DKIM and DMARC explains the checks.

For an agent this cuts two ways. Others cannot be sure a message from your agent is real unless its domain authenticates. And your agent cannot trust a sender because the name looks right.

What makes an agent's address trustworthy?

An agent's address earns trust from what stands behind it.

  • A domain that authenticates. SPF, DKIM and DMARC are published, so receivers can check that mail from the agent is real. An address on a domain you own ties the agent to your business.
  • A human who answers for it. Every agent traces to one accountable human who can review what it holds and switch it off.
  • Narrow credentials. A key limited to the agent's own inbox, with reading apart from sending.
  • A record. Each send, each release and each change of setting is written down with who did it.
  • One agent, one address. A new agent gets a new inbox. Addresses are cheap, and untangling a shared one is not.

How Trotta Mail handles agent identity

On Trotta Mail every inbox is marked agent or human, and the two are treated differently by design.

  • An agent signs up for its own inbox with one call and gets a claim_url. A human opens it to take responsibility. If the agent named a human at sign-up, nobody else can claim it.
  • After the claim the agent's key reads and sends for its own inbox. It never reaches the human's mailbox.
  • In a human's mailbox login mail starts sealed. To any API key, a sealed message with a code, a reset link or a sign-in link does not exist. A key there starts with drafts only, and the human sends.
  • A key can be limited to one inbox. A draft a key wrote is marked as written by an agent.
  • Your own domain is available on paid plans, with the DNS records checked for you. See connect a domain.
  • The audit log records keys created and revoked, agents claimed, and held mail released.
An agent gets its own address and names its human
curl -X POST https://api.trotta.email/v1/agents/signup \
  -H "Content-Type: application/json" \
  -d '{"name": "riley", "human_email": "jordan@trotta.email"}'

Mail to every inbox is screened before delivery, and each verdict carries the SPF, DKIM and DMARC results. The guard checks for spoofed senders and lookalike domains before the agent reads a message, not after.

No filter catches everything, which is why uncertain mail waits for a human. The security model has the full rules, and custom domain email for agents covers the domain setup.

To set an agent up with its own address, follow give an AI agent an email address.

Questions

Can an AI agent have its own email address?

Yes. An agent inbox gives an agent a real address it works through an API. On Trotta Mail an agent can create one itself with a single call, then hand its human a link to claim it.

Does an email address prove an agent is who it says it is?

No. An address proves control of a mailbox, and SPF, DKIM and DMARC prove the sending domain authorized a message. Neither proves which agent or human wrote it.

Should each agent have a separate address?

Yes. One address for each agent keeps mail, keys and records apart, so a problem with one agent stops there and can be traced.

Can an agent use an address at my company's domain?

Yes. On Trotta Mail's paid plans you add a domain you own, publish its records, and create agent inboxes on it. Developer includes up to 3 domains and Startup up to 25.

Is it safe to connect an assistant to my own mailbox?

It is safer when the assistant can only draft, cannot see login codes, and can be disconnected at once. In a human's mailbox on Trotta Mail, a key starts with drafts only and login mail sealed.

Sources

Give your agent an inbox of its own. Free for two inboxes. No card.