An agent needs an address

Ask how an agent should prove who it is and you will hear about new protocols, credential formats and working groups. Most of that work is real and some of it will matter. But an agent that has to deal with a supplier, a customer or a regulator this week needs something that already works everywhere, is already checked by the other side, and already leaves a record. That is an email address on a domain its owner controls.

Email is already an identity system

An address like accounts@example.com says three things before anyone reads the message. It names an organisation, because only whoever controls example.com can create it. It can be authenticated, because the domain publishes in DNS which servers may send for it (SPF), signs outgoing mail with a key whose public half sits in DNS (DKIM), and tells receivers what to do when either check fails (DMARC). And it can be answered, by any person or system in the world, without either side signing up to anything new.

For years those checks were optional in practice. They no longer are.

The large mailbox providers now enforce it

Google and Yahoo moved first. From 1 February 2024, Gmail has required every sender to personal Gmail accounts to authenticate with SPF or DKIM, use valid forward and reverse DNS, send over TLS and keep spam rates reported in Postmaster Tools below 0.3 per cent. Senders of more than 5,000 messages a day must publish SPF, DKIM and a DMARC policy (a policy of none is enough), align the visible From domain with SPF or DKIM, and support one-click unsubscribe for marketing mail. Google recommends keeping spam rates below 0.1 per cent. The same rules apply to Google Workspace senders writing to personal Gmail accounts.

Yahoo set out matching requirements and began enforcing them in February 2024: SPF or DKIM for all senders; SPF, DKIM, a passing DMARC policy of at least p=none and From-domain alignment for bulk senders; one-click unsubscribe honoured within two days; and complaints under 0.3 per cent.

Microsoft followed. In a 2 April 2025 post, it told domains sending more than 5,000 messages a day to Outlook.com, Hotmail and Live addresses that SPF, DKIM and DMARC would be required. Enforcement began on 5 May 2025, and an update on 29 April 2025 said non-compliant bulk mail would be rejected rather than sent to junk.

The three largest consumer mailbox providers now share one baseline. A domain that meets it has proved, in public DNS, that it controls its mail. A domain that does not is increasingly unable to reach anyone.

Why an agent needs its own mailbox

Many agents send mail today through a shared service account, a person's delegated mailbox, or a transactional API that sends but never receives. Each choice causes a specific problem.

A shared address hides which agent did what. When five agents send as ops@, no one can later tell which one promised the refund. A delegated personal mailbox mixes the agent's actions with a person's own, so an audit has to unpick them, and the agent inherits access it was never meant to have. A send-only API breaks the conversation: the supplier's reply goes nowhere, or to a person who was never told what the agent said.

An agent with its own mailbox on the organisation's verified domain avoids all three. Its messages are attributable to it. Its access is scoped to it. Replies come back to it, so it can carry the thread to the end. And it can be switched off without touching anyone else's mail.

The record is the point

The mailbox matters as much for what it keeps as for what it sends. Email is one of the few systems where both sides of a conversation hold a copy, with timestamps and authenticated headers, in a format auditors and regulators already know how to read.

For an agent, that record answers the questions that come after the work: what did the agent tell the customer, when did the supplier confirm the price, who approved the exception. It answers them only if the record is complete, searchable and readable by a person without special tools. A trace in a vendor's observability console is useful to an engineer. It is not something a finance director or an ombudsman will read.

How this compares with emerging agent identity standards

Work on agent-specific identity is moving, and it is worth tracking.

The OpenID Foundation published Identity Management for Agentic AI on 7 October 2025. It finds that current authentication and authorisation standards can secure simpler agents working inside one organisation, but were not designed for recursive delegation chains or trust that has to cross organisational boundaries. It recommends that enterprises manage agents as identities within their existing identity infrastructure, with lifecycle management and accountability, and that standards bodies formalise delegation quickly.

In the US, NIST's National Cybersecurity Center of Excellence published a concept paper on 5 February 2026 on software and AI agent identity and authorisation. It sought input on identifying, authorising, auditing and ensuring non-repudiation for agents; comments closed on 2 April 2026.

Payment-specific identity is arriving through commerce. Google's AP2 uses signed mandates as proof of what a user authorised. Visa's Trusted Agent Protocol, introduced in October 2025, helps merchants tell legitimate agents from malicious bots.

In the EU, each Member State must make at least one EU Digital Identity Wallet available to citizens and residents by the end of 2026. That is identity for people, and the people who own and direct agents may use it to prove who they are.

None of these replaces an address. They answer narrower questions: is this token valid for this API, did this user approve this cart, is this bot welcome on this site. Email answers the general one, can this party be reached and held to what it said, across every organisation at once. The sensible design uses both: a verified mailbox as the agent's public address and record, and newer credentials where a specific counterparty asks for them.

What this means for CTOs and executives

For CTOs: give each agent its own mailbox on a domain you control. Meet the Google, Yahoo and Microsoft baseline on that domain, with SPF, DKIM, an aligned DMARC policy, and a plan to move from p=none to enforcement. Keep agent mail in the same retention and search system as staff mail, so one query covers both. Treat agent identity work from OpenID and NIST as something to adopt as it matures, not something to wait for.

For executives: an agent that writes to customers or suppliers is speaking for the business. Insist that it does so from an address the business owns, that every message is kept, and that a person can find and read the full exchange without asking an engineer.

Where Praecise fits

Praecise Inbox gives every agent and person an address and a searchable record: mailboxes on a domain the business verified once, with everything said kept in one record a person can open and search. On Praecise Cloud, each agent gets an inbox and a way to pay alongside its compute. Ipnops connects payments, accounting, commerce and mail and acts across them inside limits the owner sets, and every run leaves a record.