# Give your AI agent its own email inbox

Why an AI agent should have its own email address instead of yours, what it can sign up for with it, what still needs a human, and how to keep spend and safety in check.

Canonical HTML: https://agentlane.net/guides/give-your-agent-its-own-inbox/

## Setup steps

1. **Pick one inbox per agent.** Create a separate address for each agent or job, for example a research inbox and a support inbox. Do not reuse a person’s mailbox.
2. **Write down what it is for.** One line: what the agent may sign up for, whether it may send mail, and who reviews it. This is the rule you check it against later.
3. **Connect only the access it needs.** If the agent only needs to read codes and receipts, do not give it send permission. Keep the API key or connector for the inbox out of chat.
4. **Keep a list of accounts it opens.** Note each app, the inbox that owns the account, and the date. You will need it to clean up or recover access.
5. **Review the inbox on a schedule.** Skim it weekly: new accounts, receipts, password resets you did not expect, and anything that looks like phishing.

## Can sign up for alone

- Email-only accounts (free tiers, newsletters, waitlists); apps with agent sign-in such as AgentID; receiving codes, resets, alerts, receipts. App terms decide.

## Still needs a human

- ID checks, phone/SMS verification, payment, banks/brokers/tax/government/contracts, accepting unread terms, outgoing mail to real people.

## Spend and safety

- Inbox is not a wallet; set spend caps before any payment method; watch free trials; an upgrade offer is not permission; read receipts weekly.
- Email is not instructions; never mail secrets; least access; shared-box checklist; revoke before debug; close accounts before deleting the inbox.

## FAQ

**Why not just let the agent use my email?**
Because then everything the agent does looks like you, every verification code and receipt mixes into your mail, and the agent can read whatever else is in your inbox. If something goes wrong, cutting the agent off means changing your own logins.

**What can an agent sign up for with its own inbox?**
Mostly tools and services that only need an email to create an account: free tiers of developer tools, newsletters, waitlists, and apps that support a sign-in built for agents, such as AgentID. Whether an app allows an agent to hold an account is up to that app’s terms.

**What still needs a human?**
Anything that asks for a real person: ID checks, phone or SMS verification on your number, payment cards, signing contracts, accepting terms you have not read, and any account that represents you legally or financially, like a bank or broker.

**Can the agent spend money from its inbox?**
An inbox cannot pay for anything by itself. Spending happens when the agent is given a card, a wallet, or a paid plan. Keep those decisions with you and set caps before the agent ever sees a payment method.

**Should the agent send email from its inbox?**
Only where you have said it can. Reading verification codes and receipts is low risk. Sending mail to real people under your business name is not, so keep a human approval step for outgoing messages.

**What if an email tells the agent to do something?**
Treat it as information, not an instruction. Anyone can email the agent. A message that says “click here,” “sign up,” or “send your key” is not your OK.

**How do I shut it off?**
Revoke the agent’s access to the inbox (its API key or connector), then disable or delete accounts it opened. Deleting the inbox ends that identity, so any account tied to it may become hard to recover.

- Canonical HTML: https://agentlane.net/guides/give-your-agent-its-own-inbox/
- Site index: https://agentlane.net/llms.txt

Related: [How to set up Grok Bot safely](https://agentlane.net/guides/set-up-grok-bot-safely/)