Is Grok Bot Safe? Approvals, Vault and Access to Your Computer
- Grok Bot has real controls: approval cards, Auto Review, Vault for logins, and a setting that keeps Bots off your own computer.
- It also has limits the official docs state plainly, including a shared computer for all your Bots and checks that lower risk without removing it.
- What it means for you depends on what you connect and which approvals you set. The checklist near the end takes about ten minutes.
If you're asking is Grok Bot safe, this page won't give you a yes or no. It shows you what a Bot can reach, who decides each step, and which settings to change before you connect your email.
I'm not part of the company that makes Grok Bot, so read the official security pages before you connect anything.
Is Grok Bot safe to use?
It has real controls, and it also has limits the official docs state plainly. How much risk you take on depends on what you connect, whether a Bot can touch your own computer, and which actions you keep behind an approval, so the answer is partly yours to set.
The official how-it-works page sums up the trade-off well. A Bot has no account or credentials of its own, so it can't do anything you couldn't, and every action traces back to you. App sign-ins stay on Cursor's servers, and Bots use apps without getting the sign-ins. But the page is also honest that its checks "lower the risk from harmful content, but they don't remove it."
What a Bot can reach, and who decides
| What | Can a Bot reach it? | Who decides |
|---|---|---|
| Its cloud computer | Yes. Bots work on a cloud computer with a browser, files and a terminal | Auto Review checks each action; sensitive ones stop for your approval |
| Your own computer | Not by default | You, in Settings > Computer > Your computer (Ask every time is the default) |
| Your apps | Only the apps you connect | You choose which apps to connect; sign-ins stay on Cursor's servers |
| Your passwords | Filled in from Vault, a password manager such as 1Password, or a Secure Form, without being shown to the model | You choose what to save |
| Your cards | A card saved in Vault can be seen and typed at checkout in Grok Bot app chats, without a separate approval each time | You, by choosing whether to save the card |
| Sending email | From a chat request, after Auto Review allows it; from a routine or work you didn't start, it waits as a draft | Auto Review, plus you for drafts, forwards and some attachments |
What can a Bot do without asking you?
Most actions run on their own after Auto Review checks them, and when one needs your OK, the Bot shows an approval card and waits, sometimes with a warning that the action can't be undone. Only your own messages count as instructions; text in an email or on a web page doesn't.
Auto Review is on by default. It checks each action a Bot is about to take. The official approvals page says it "is model-based," and that it should sit alongside least privilege and clear approval limits rather than replace them. With Auto Review off, the Bot runs actions without checking them first.
Email has its own rules. Email a Bot sends from its own address after you ask in a chat goes out once Auto Review allows it. Email from a routine, an incoming message or other work you didn't start waits as a draft until you press Send. A forward, or an email with an attachment Auto Review can't fully read, always waits for your OK.
My own templates add a stricter rule on top. The Daily Email and Calendar Brief, Agency Reports and Admin Helper, Insurance Lead Follow-Up Writer, New Insurance Lead Alert, and Speed-To-Lead Bot for Insurance Agents are draft only. The Daily Brief may send only to you; it never sends to anyone else. The Subscription and Bill Cutter drafts the cancellation and sends it only after two yeses: one on the item and a second on the exact text.
For what Bots can actually do day to day, see What Can Grok Bot Actually Do? Real Use Cases for Business Owners.
Does it access my own computer?
Not by default: Bots work on their own cloud computer, and working on your own machine is a separate setting under Settings > Computer > Your computer > Execution on this computer. Ask every time is the default, and the other choices are Always allow and Never allow.
The official docs advise Never allow unless a Bot has a specific reason to work on your local files. A team admin can also cap this setting for everyone on the team.
Does it work when my computer is off?
Yes: Bots work on their own cloud computer, so the work keeps going while your laptop is closed. Closing the app, your laptop or your phone doesn't stop a background turn or a routine, so pause a routine or tell the Bot to stop if you want the work to end.
That's useful for scheduled jobs, and it's also a reason to think about what a routine is allowed to do while you're away. A routine tells one Bot when to run a job, on a schedule or when something happens, and it runs in the cloud. Email from a routine waits as a draft until you press Send.
What happens to my passwords and cards?
Bots sign in without seeing your passwords: they use a login saved in Vault first, then a connected password manager such as 1Password, then a Secure Form, and those values are filled in without being shown to the model. Cards are different, because a saved card fills in at checkout without a separate approval.
- Passwords. Logins from Vault and values from Secure Forms are filled in without being shown to the model.
- Hard sign-ins. For single sign-on, passkeys, CAPTCHAs and similar steps, the Bot hands you the computer with Take over.
- Never in chat. Don't send a password, a one-time code or an API key in an ordinary chat message.
- Cards. In chats in the Grok Bot app, a Bot can see the number and security code of a card you saved in Vault, so it can type them at checkout. A saved card fills in without a separate approval each time.
- To approve every payment, don't save the card.
What are the real limits?
All your Bots share one computer and aren't a security boundary from each other, the official checks lower risk without removing it, Auto Review is model-based, and sharing a Bot exposes its setup. None of this is hidden; the official docs say each of these plainly.
- One shared computer. All of one user's Bots share one cloud computer, including files, browser sessions and command line credentials. The docs say not to use separate Bots as a security boundary. Sign out of services a Bot no longer needs, and delete sensitive files when the work is done.
- Harmful content. Text in an email or on a web page doesn't count as an instruction, and the checks lower that risk. They don't remove it, so keep Ask first rules on anything you can't undo.
- Auto Review is a model. It's model-based, so treat it as one layer alongside your own approval rules and a least-privilege setup.
- Sharing exposes setup. Sharing a Bot as a template exposes its configuration (identity, description, skills and routines) to anyone with the link. Remove secrets and customer data first.
- Installing templates. The official template guide advises evaluating a template the way you'd evaluate software: read its context and integrations, and make sure everything is what you expect before you install it.
None of the sources used for this page name certifications or audit claims, so none are claimed here.
What should you set up first?
Before you connect email, set Ask first rules for anything you can't undo, keep Bots off your own computer, decide whether to save cards, and connect only the apps a job needs. The official approvals page calls this a least-privilege setup: start with read-only tasks and drafts.
Set these first
- Ask first rules for sending, publishing, purchasing and deleting.
- Execution on this computer set to Never allow, unless a Bot needs your local files.
- Vault choices. Save logins you want Bots to use; leave cards unsaved if you want to approve every payment.
- Least-privilege apps. Connect only the tools a workflow needs, and start with read-only tasks and drafts.
- Leave Auto Review on. With it off, the Bot runs actions without checking them first.
- Regular reviews. Look over your connected apps and routines from time to time, and clean sensitive files off the shared computer.
My own rule for any bot that touches the outside world: drafts only, a person approves, and for anything that sends or cancels, one yes on the item and a second yes on the exact text and recipient.
If approval cards start getting in your way, Grok Bot Not Working? Fixes for Routines and Permission Prompts shows how to cut them back with narrow rules instead of switching checks off. For which apps are worth connecting at all, see What Apps Can Grok Bot Connect To?.
FAQ: is Grok Bot safe?
Can a Bot send email without me?
It depends on where the request came from. Email you ask for in a chat goes out once Auto Review allows it, while email from a routine or work you didn't start waits as a draft until you press Send. Forwards, and attachments Auto Review can't fully read, always wait for your OK.
Can it see my passwords?
Logins from Vault and values from Secure Forms are filled in without being shown to the model. Cards are the exception: in Grok Bot app chats, a Bot can see a saved card's number and security code so it can type them at checkout.
Can it read my local files?
Not by default. Execution on this computer is set to Ask every time unless you change it, and the official docs advise Never allow unless a Bot has a specific reason to work on your local files.
Does it train on my data?
According to the official approvals page, training opt-out follows your Cursor account and privacy settings. Grok Bot requires cloud data storage and doesn't support Legacy Privacy Mode.
What happens when I delete a Bot?
The official Grok Bot FAQ says deleting your account removes your Cursor account and Grok Bot access, Grok Bot agents, chats, computers and connected apps, and Cursor chat history and settings, and that all data is removed within 30 days. Files on your own computer stay there. For a single Bot, the official Create and manage Bots docs say Delete erases that Bot's profile, conversations, memory, routines and attachments, including shared facts it saved for your other Bots, and any template you shared from it. Files and sign-ins on the shared computer can remain. Backend retention follows the applicable Cursor terms. Hide the Bot instead if you may need its work later.
Can my team admin limit it?
Yes, in part. A team admin can cap the Execution on this computer setting for the team. For other team controls, check the official security page.
If you're new to the basics, What Is a Grok Bot? A Plain Guide for Business Owners explains how Bots work. If you want a set of ready-to-install bots, in order, that look for money before they ask you to spend more of it, see The Payback Roadmap.
This bot is one piece of a full AI team. See how I run my business with a team of Grok bots. The AI Chief of Staff: How I Run My Business With a Team of Grok Bots