Grok Bot Approvals and Auto-review: Cards, Rules and Expiry

Auto-review is a check that runs before your Bot acts. It is on by default, and for each action it either lets the action run or has the Bot stop and ask you. When it asks, you get an approval card in the chat and the Bot waits for your answer.

Cursor's security docs describe it as an independent review model that evaluates risky Bot actions before they run. The list of what it looks at is longer than most people expect:

• Shell commands
• App calls, meaning actions in apps you connected
• Computer use
• Automation writes, which are changes to routines and event triggers
• Delegation, such as launching a Cloud Agent or a subagent

For any of these it can let the action proceed, require approval, or deny it. Your Custom Rules can change that choice for the actions you care about, and built-in safety checks apply whatever your rules say.

It has gaps, and the docs name them. Auto-review does not review every side effect: memory writes and most settings changes are the examples given. Cursor positions it as a complement to explicit boundaries and least privilege, working alongside controls that do not depend on a model's judgment, such as per-action approvals and per-user isolation. The security docs call the request itself the strongest boundary, so say what the Bot may change and where it must stop, as in their example: "Don't change the campaign or message the agency. Ask for approval after showing the current value, proposed value, and expected impact."

The Bot shows a card when Auto-review decides an action needs your OK. Any action that matches one of your Ask first rules always gets a card, regardless of what the reviewer would otherwise decide. The card says Review an action.

Auto-review takes into account what you asked for in your own messages, and only those. Text that came from somewhere else, an email the Bot read or a web page it opened, does not count as your request. So an instruction sitting in a page or an email can't stand in for something you asked for.

Some cards carry a warning line, such as This can't be undone, This changes sign-in or security settings or This shares your data outside this chat. Those are worth slowing down for.

Email the Bot sends from its own address follows its own pattern, which the table below sets out. The full detail sits on the agent email page.

Most cards show three buttons, and the difference between the first two is whether a rule gets saved. The labels change a little depending on what raised the card, which the second half of the table covers.

On the phone, tap a card to see the full request and all of its buttons. The security docs say the controls are the same on iPhone.

An approval controls the proposed action and nothing before it. It does not undo work the Bot already finished, and Cursor's guidance is blunt that nobody should approve an action whose target or effect they can't identify. Always allow is, I think, the button to be most careful with, because the rule it saves will apply to actions you have not seen yet.

It depends on how the work started. When the work began from a chat message, the card waits for an answer and does not expire on a timer. When a routine, an incoming message, a trigger, a message from another Bot or other background work raised the card, it expires after 30 minutes.

Two other things close an open card. Sending a new message in that chat instead of answering closes any open card there, and so does changing your Auto-review settings.

A closed card shows Expired and says the Bot didn't do the action. If the card offered a rule, you can still press Always allow next time, and the Bot saves the rule so it can go ahead next time. Otherwise, ask the Bot to try the task again.

Expiry only hits work that started without you, so catching cards in time is mostly about seeing them. A pending approval appears in three places:

• In that Bot's chat, as the card itself.
• In the sidebar, where the Bot shows that it needs attention.
• As a notification, if the Bot's Notifications switch is on. Desktop alerts don't show while Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. is the window you're using, though the sidebar still does.

For Bots that run routines, turn on Notifications. On the phone the switch is in the Bot's profile, and on desktop you ask the Bot to turn its notifications on. Phone alerts also need notifications allowed for Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. in your phone settings.

Custom Rules let you decide ahead of time how Auto-review treats a specific action. On desktop, open Settings > General and go to the Bot section, where Custom Rules appears under the Auto-review switch while Auto-review is on. On the phone, open Settings, find the Bot group and tap Custom Rules. A card paused by one of your rules also has Edit rule, which takes you straight there.

1. 1Under When Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. wants to:, describe one action in plain words, like "reply to emails for me".
2. 2Under It should:, choose Allow automatically or Ask first.
3. 3Press Add Rule.

Write one short rule per action. If two rules conflict, Ask first wins. The two choices are less symmetric than they look, though. An Ask first rule always stops a matching action, while an Allow automatically rule lets a matching action proceed only when the reviewer finds no other reason to stop it.

The security docs suggest keeping rules narrow and tied to a known action, such as "ask first before sending any external email" or "allow automatically when running git status in /workspace/reports", and avoiding broad ones like "allow everything in the browser". Rules are saved to your account, so they apply from every desktop and the mobile app.

That last row catches people out. Once you are at 20 Allow rules, every Always allow quietly retires an older one, so a rule you relied on can disappear. To edit a rule on desktop, use the edit button next to it and press Save Rule, or Cancel to leave it as it was, and use the delete button to remove it. On the phone, tap a rule and choose Save Rule or Delete Rule.

If you're on a team, your admin can set Auto-review for you. The security docs mark both admin controls, enforcing Auto-review and adding team Auto-review rules, as Enterprise only. What you see as a member is this:

• If your admin requires Auto-review, the switch stays on and shows Required by your admin. You can't turn this off.
• Team rules show Required by your admin. You can't edit or delete this rule. Team rules always apply. Your own rules apply only to you.
• Your team can also share rules that apply to your Bots instead of your own rules. Built-in safety checks still apply.

Personal rules on top of team rules can only make behavior stricter. If an admin turns enforcement off, team rules stop applying and you go back to your own rules only. A group you belong to can also lift the lock. The admin side of all this, including where the switch lives on the dashboard, is covered on Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. security and privacy.

When nothing is enforced, you can turn Auto-review off. On desktop, open Settings > General and turn off the Auto-review switch in the Bot section. On the phone, turn off Auto-review in the Bot group of Settings.

The rule across every shared setting is that the person the Bot is answering is the person who approves. Nobody else in the room can answer the card for them.

A Team Bot also asks before it uses your personal accounts. How Team Bots are shared and whose apps they use is on the Team Bots page.

Three kinds of card can appear in a chat and they look alike at a glance, which is probably why the help pages keep separating them. An approval card asks whether the Bot may take an action. The other two collect something from you.

• A Secure Form collects a password, key or other secret and saves it to the Bot's vault. See Grok BotAn early AI teammate product from SpaceXAI and Cursor that works across apps on a persistent cloud computer and asks for approval on selected actions; launch materials describe it as beta, but Cursor says Beta Services terms do not apply. Press Enter for the full definition. Vault and secrets.
• A connect card asks you to sign in to an app so the Bot can use it.
• A page that asks you to type a password is a different prompt again. Use the secure card for secrets, never the chat.