Should You Connect Meta Muse to Gmail?

Should you connect Meta Muse to your Gmail?

No. Give Muse its own email address with AgentMail instead. Gmail was built for one person clicking through one inbox. Muse is an agent that runs around the clock on a cloud VM, logs in from Meta's servers, sends on a schedule, and keeps working after you close the app. That is exactly the pattern Google's systems are built to flag, and it is the same pattern that has gotten agent-connected Gmail accounts disabled on OpenClaw, Hermes, and Grok Bot. Meta doesn't trust an agent with your inbox either: Muse's email connector strips out verification codes, reset links, and magic links so the agent can't act as you. Which means Muse can't sign up for anything through your Gmail, and every email it does send goes out under your name. AgentMail gives Muse a real inbox it owns. It signs up, sends, and receives as itself, and your personal Gmail stays out of the blast radius.

On this page

• Why isn't Gmail built for agents like Muse?
• Hasn't this happened before?
• What does Meta's own design tell you?
• What happens when Muse sends as you?
• Why does a personal agent need its own identity?
• What if you don't want to connect Gmail at all?
• Can't Muse just make itself a Gmail account?
• How do the two compare?
• How do you give Muse its own email address?

Why isn't Gmail built for agents like Muse?

Gmail was designed in 2004 for a person sitting at a laptop. Everything Google does to protect an account assumes that person: the device they usually log in from, the hours they're awake, the way they write and send. When activity stops looking like that person, Google's systems treat it as a possible takeover or as spam.

Muse breaks every one of those assumptions. It doesn't run on your phone. It runs on a Linux VM in Meta's cloud, with its own browser, terminal, and scheduled jobs. It keeps going after you close the app. Early reviewers set it up to triage their inboxes at 5am and chase unanswered mail at 6pm, every day. That is a login from a server you've never used, activity at hours you're asleep, and sends on a fixed schedule. To Google, that doesn't look like you. It looks like someone else in your account.

Testers saw Google react within a day. One asked Muse to order from DoorDash and sign in with Google. Google sent a "someone is trying to sign in" prompt to his phone. He confirmed it. Muse tried again, Google challenged again, and the loop repeated three times before he gave up. Google didn't recognize Muse as him, because it isn't him.

Hasn't this happened before?

Every time. Each new agent platform goes through the same cycle: people connect their personal Gmail because it's the account they already have, the agent starts acting on it, and Google starts disabling accounts.

It happened with OpenClaw agents wired to Gmail over OAuth and IMAP. It happened with Hermes agents. In August it happened with Grok Bot, where accounts created and run by the bot were disabled within days, with Google's notice saying the account "might have been created by a computer program or bot." We wrote up each wave: OpenClaw, Hermes, Grok Bot, and what to do once it happens to you. Appeals rarely work. When the account goes, so does every integration that depended on it.

It happened to me, too. I gave a small agent API access to my personal Gmail and told it to sign up for Reddit. Google disabled the account that afternoon. Not a test account. My first email address, with fifteen years of threads, photos, and receipts in it.

Muse is the same kind of agent, running the same kind of schedule. The pattern won't skip it.

What does Meta's own design tell you?

That Meta doesn't think an agent should act through your inbox either.

Meta's security post for Muse says the email connector filters out one-time passcodes, password reset links, and login magic links, using fixed rules plus a classifier. The reasoning: your inbox can reset every account you own, so an agent with access to it could be steered into taking them over. Meta looked at the same risk we've been writing about and decided the agent should never see those emails.

The consequence is that Muse can't complete an email verification through your Gmail. Any signup that sends a code to your address stalls. Testers hit this on day one: one asked Muse to book an Airbnb, and it got as far as payment before explaining it couldn't finish because Airbnb requires an account, then asked for his login.

So connecting Gmail gives you the ban risk of an agent in your inbox without the thing you'd want an agent in your inbox for. Muse can read your mail. It can't use it to get anything done on a new site.

What happens when Muse sends as you?

Every email Muse sends through your Gmail goes out under your name, from your address, on your reputation.

A quote request to five moving companies. A follow-up to a car dealer. A complaint to an airline. Each one puts your personal address on another vendor's list, and every reply lands in your inbox mixed in with everything else. If one message trips a spam filter, or one recipient marks it as spam, that complaint is scored against your address, not the agent's.

And when something goes wrong, the damage isn't limited to the task. For most people, their personal Gmail is the recovery address for their bank, Apple ID, work accounts, and kids' school portal. Hand it to an agent and all of those share the agent's blast radius.

Why does a personal agent need its own identity?

Most errands a personal agent runs start with an email address. Creating an account, confirming a booking, asking a business a question, following up when nobody answers. When the only address available is yours, the agent has two options: act as you, which Meta blocks and Google punishes, or stop and ask you to do it yourself.

A separate inbox gives the agent a third option. Muse signs up for the restaurant platform, the retailer's account, or the waitlist with its own address. The verification code lands in an inbox Muse is allowed to read. It finishes the job. Your Gmail, and every account that can be reset through it, stays out of it.

Instinct, the $2.5B personal agent startup, shipped this on September 8, the same day Muse launched. Every Instinct agent now gets its own address. Founder Noah Shinn's examples: contacting a restaurant about a special request, asking a business about availability, and signing up for a service the agent needs to finish a task. His stated reason was that most everyday tasks run through email, and the agent shouldn't clutter the user's inbox doing them.

An address also makes Muse reachable. With a borrowed Gmail, nobody can CC Muse on a thread or email it directly, because it has no address to CC. Give it one, and a contractor can reply to Muse, your partner can loop it into the wedding-venue thread, and another person's agent can write to yours. Zuckerberg has said Meta already has agents interacting with each other internally, and that multiplayer is where Muse is headed.

Muse doesn't ship with an address of its own. AgentMail fills that gap.

What if you don't want to connect Gmail at all?

The loudest reaction to Muse on Hacker News wasn't about features. It was some version of "I want this product, and I don't trust Meta with my inbox."

An agent-owned address means you never have to. Skip the Gmail connector. Forward Muse only the emails it needs: the order confirmation for the return, the school newsletter with the field trip form, the thread with the contractor. Or CC Muse's address on a group thread so it can follow along. Muse works from what you sent and nothing else. Your Gmail stays disconnected, and you can shut off Muse's inbox whenever you want.

This is the least-privilege way to run a personal agent. Muse sees only what you hand it.

Can't Muse just make itself a Gmail account?

It can try. Muse has a real browser, and nothing stops you from asking it to go to Google's signup page. Don't.

Google's account policy is short: "Don't use programs (called bots) to create fake accounts." There is no volume threshold. That is the exact setup that got Grok Bot users' accounts disabled in August. A Gmail account an agent creates for itself is the one email setup with the clearest record of getting shut down.

AgentMail inboxes are made to be created by software. Muse, or you, can create one with a single API call, and nothing about that breaks anyone's terms.

How do the two compare?

How do you give Muse its own email address?

Tell Muse to get one. Setup takes about five minutes and needs no code.

1. Ask. Message Muse: "Get yourself an AgentMail email address." It replies with a plan.
2. Create an API key. Sign up at the AgentMail console, create an API key, name it "Muse", and copy it.
3. Paste the key into the Secrets tab, not the chat. Muse opens the tab for you.
4. Approve the connection. Muse's request to share information with AgentMail appears in the approval section on the right of the app. Approve it there. You can grant a standing permission for the site so it doesn't ask on every call.
5. Test it. Muse creates an inbox, choosing the address and display name itself unless you name one. Ask it to email your personal address. When the message lands, you're done.

Full install steps and a recipe library of copy-paste prompts are on /build/muse. The step-by-step walkthrough is in How to Give Meta Muse Its Own Email Address.