Grok Bot Now Comes Bundled in The X Subscription. It Still Shares One Computer, and It Still Won't Say Which Model Runs It.

In August we wrote that Grok Bot wants your passwords, because xAI's always-on agent ran every one of your Bots on a single cloud computer with a single set of logins, and the docs said not to treat separate Bots as separate security boundaries. Two months, one acquisition, and one corporate rename later, Grok Bot has a new price, a new name on the door, and a new set of surfaces. The security sentence is still in the docs, word for word.

Today, October 12, 2026, SpaceXSI announced The X Subscription: one plan that replaces SuperGrok, Cursor's plans, and X Premium. Five tiers, Grok Bot on the middle three. That is the news. Around it sit three weeks of Grok Bot releases that deserve a ledger, because the Bot you are paying for on October 12 is not the Bot we reviewed on August 15.

We read the subscription post, the pricing and unification docs, the Team Bots post and docs, every changelog entry from v0.61 to v0.70.2, the approvals-and-security pages, the September 22 customer-support case study, and the X posts that stand in for primary sources where SpaceXSI has published none. Here is what held up.

The X Subscription, With the Blank Left Blank

Premium is the headline tier. SpaceXSI's description: "The benefits of SuperGrok, Cursor Pro, and X Premium+, previously $90/month combined." Thirty dollars for what used to be three subscriptions is a real cut if you used all three; the pricing docs add team seats at $30, $100, and $200 per seat for the three paid tiers with Bots.

The multiples have no denominator. Premium is "4x Plus Plan," Super "15x," Ultra "50x," and the Plus plan's "Base" usage is never quantified anywhere on x.ai or docs.x.ai. Grok Bot "usage draws from the weekly shared pool across Grok, Grok Bot, and Cursor," per the docs, and no page states a per-Bot price, a Bot count, or compute hours. We print a dash for all three.

Switching is one way. The unification docs: "A completed unification is permanent. It can't be undone, and Support can't undo it for you." Your old Cursor, SuperGrok, and X Premium subscriptions are cancelled at checkout with a prorated refund. SpaceXSI says "Nothing changes for existing SuperGrok, Cursor, or X Premium subscribers"; the old plans have vanished from the pricing pages, and whether they are closed to new sign-ups is something no SpaceXSI page states, so we do not state it either.

The Tesla tier is unmapped. On September 22 Tesla switched on Grok Bot in its cars, "Currently available to SuperGrok Heavy subscribers, with more tiers to follow." SuperGrok Heavy was the $300 plan. Which X Subscription tier inherits it is not published.

Three Weeks of Changelog, Dated

House rules: the changelog is the primary source; press dates are wrong more often than you would expect, and two of ours were too.

• September 22: x.ai published "How SpaceXSI is using Grok Bot to scale customer support." It is a first-person internal case study, not a product launch: a 175% rise in support tickets after the Cursor merger "but we have not had to hire any new people," resolutions "for as low as $0.20 to $0.30" against "a flat $1 to $4 per resolution" for traditional tools, 99% of refund requests resolved without a human, 20,000 product-feedback points a day. Starlink and Tesla are not on the page; the 15,000 Starlink calls a day figure in some coverage is attributed to Grok Voice by one outlet and appears on no x.ai page. Separately, Bloomberg reported Grok Bot at 418,000 weekly users as of September 14, from a presentation at a London event.
• September 26, v0.61.0: Team Bots in the app and in Slack.
• September 28: Team Bots public beta "on Teams and Enterprise plans." You build one Bot around a shared role from Context (files, instructions, skills), Plugins, Credentials, and Memories; "Although the Bot is shared, each person's conversations with it remain private"; usage "counts against that teammate's own Grok Bot allowance, not the owner's." Secrets are encrypted at rest and "the value is replaced with [REDACTED]" in command output, files, and plugin calls. Templates are Customer Bot, EPD Teammate, Marketing Bot, and Data Bot. No price.
• September 28, v0.63.0: In Slack threads "a Bot that needs your approval posts a Review an action card only you can answer, instead of dropping the action." A Team Bot in Slack can message one of your own Bots, with the reply coming to you privately. That is narrower than the "Bot-to-Bot messaging" some write-ups described.
• September 29 and 30, v0.64.0 and v0.65.0: Team Bot managers; voice chat with a Team Bot; automatic updates on by default.
• October 2, v0.66.0 and v0.67.0: "Your primary Bot, marked with a star, will now spot work it can take off your plate and offer to handle it." The marketplace became Connect Apps.
• October 7, v0.68.1: Bots build slide decks, draft formatted email, and search your Gmail spam folder.
• October 11, v0.70.2: "Bots can claim their own email address and receive mail there, and email they send from it waits for your approval on a draft card." Bots can search X. In-chat apps and payments.

A proactive agent with its own inbox, its own Slack handle, and your Gmail is a different product from the August one. Which brings us back to the security page.

What Did Not Change

The approvals-and-security docs, read today: "All of your Bots share one cloud computer assigned to your user account. Files, browser sessions, and command line credentials on that computer are available across your Bot roster." And, unchanged since our August review: "Do not use separate Bots as a security boundary."

The mitigations are real and worth listing. Passwords, passkeys, two-factor codes, CAPTCHAs, and payment confirmations are supposed to be handed to you ("the Bot should hand you control of the computer"). A secure secret request masks the value, excludes it from the transcript, and hides it from the model. Connector OAuth tokens stay on Cursor's backend. Approvals gate sending messages, publishing, financial transfers, deletions, permission changes, and production changes. But the docs also say what an approval is not: "An approval controls the proposed action. It does not reverse work already completed." Auto Review, which decides what needs asking, is model-based, and enforcement of review rules and a model allowlist is Enterprise-only; Teams self-serve gets neither, nor Network Controls, SCIM, audit logs, or Action Recording.

Which model? No SpaceXSI page says. The security docs: "Cursor manages model selection. There is no customer-facing model picker, and the serving mix can change over time." We print a dash where the Bot's default model would go. On October 7 Elon Musk posted that "Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs." Several outlets reported Grok Bot as already routing to Claude. As of today no x.ai news post, changelog entry, or doc confirms it is live, and the docs do not say whether third-party model calls draw from your weekly pool.

One incident, unconfirmed. On October 6 XMTP Labs CEO Shane Mac posted that "an AI agent posted my personal bank balances into our company Slack. As me," on October 1, naming Grok Bot. SpaceXSI has not confirmed it; one outlet reports a Community Note calling it fake, which we could not see. We print it as a user claim, nothing more.

The name. On September 29 an executive order directed federal agencies to say "Super Intelligence" and "SI" instead of "Artificial Intelligence" and "AI." On October 4 Musk posted "No more AI / SI / It's better" and confirmed the company would rename; on October 10 @SpaceXAI became @SpaceXSI, "SpaceX Super Intelligence." The legal entity is still SpaceXAI LLC and the docs still mix both names.

Team Bots, Read Closely

Team Bots is the feature most likely to end up holding company credentials, so the docs deserve a closer read than the launch post.

What is shared and what is not. "The plugins, secrets, skills, and files the owner adds are available in every teammate's conversation." Conversations are not: "Every conversation stays private. Each teammate gets their own chat," and notes the Bot keeps about a person are private to that person. Usage is charged to the teammate asking, not the owner. The creator can name Managers who edit the Bot; Unpublish removes it for everyone.

How secrets are handled. The docs say stored values "are encrypted when stored, and the app shows only each secret's name and description," and that the Bot "can't see a value either. It uses a secret by name, and the value is replaced with [REDACTED] in the output of the commands it runs, the files it reads, and its plugin calls." That is a materially better design than typing a password into a chat, and it is the same Bot roster that shares one computer, so a secret the Bot can use is a secret every Bot on that account can use.

Data handling. "With Privacy Mode enabled, customer data is not used for training." Deletion "follows the DPA," within 30 days. Zero Data Retention "follows Cursor's existing provider agreements," meaning the model providers behind the Bot keep no prompts or outputs; Legacy Privacy Mode is unsupported. Members sign in with their Cursor account, so existing Cursor SSO applies.

What Enterprise buys that Teams does not. Network Controls with directory groups, SCIM 2.0 provisioning, audit logs, Action Recording (off by default), enforced Auto Review rules, and a model allowlist that, in the docs' own words, has "enforcement not guaranteed." A Teams-plan Team Bot has none of those, and a proactive Primary Bot that "will now spot work it can take off your plate and offer to handle it" is on by default for everyone.

Verified vs Unconfirmed: The Scorecard

Who Should Use It

Use The X Subscription if you already pay for two of SuperGrok, Cursor, and X Premium; Premium at $30 is cheaper than the pair, and Grok Bot comes with it. Use Team Bots if your team shares a role-shaped workflow and can live with Enterprise-only enforcement of the review rules.

Skip it if you need to know what you are buying before you buy it: the usage pool has no published base, the Bot has no published model, and unification cannot be undone. Skip giving a Bot its own inbox or payments until you have read the approvals page twice, because an approval "does not reverse work already completed," and all your Bots still share one computer and one set of credentials.

The frontier is elsewhere: on always-on agents the comparison is OpenAI's Dots, which gives each agent its own cloud computer and still ships unpriced extras and a geo-fence of its own. On models, Grok 4.7 sits at 46 on the Artificial Analysis index against Claude Opus 5.5's 58, and Grok 4.8 has a parameter count from Musk and no date. Our model comparison has the current standings side by side.

Update, October 16, 2026

• October 13 and 14: Cursor's status page, where Grok Bot incidents are posted, logged routines failing to trigger on October 13 and a 5-hour-18-minute Grok Bot degradation on October 14. SpaceXSI's own status page lists none of it.
• October 14: Grok Voice 3.0 took first place on Artificial Analysis's Speech to Speech Index; Team Bot voice chat runs on the Voice stack, and SpaceXSI has not said which version.
• October 15: Builder.io's Alice Moore, after paying $20, $60, and $200, wrote that "GrokBot is the closest thing we have to a working implementation. But it's just not there yet, and the reasons are architectural," and that she "was paying for inference that was basically deciding, Hmm, I guess we shouldn't talk."
• October 16: Still no x.ai page confirming the Claude Opus 5.5 routing, no per-Bot pricing, and no mapping from SuperGrok Heavy's in-car Tesla access to the new tiers. The security sentence is unchanged.

Sources

• SpaceXSI: Introducing The X Subscription
• docs.x.ai: Grok Bot on The X Subscription
• docs.x.ai: The X Subscription pricing
• docs.x.ai: How to unify accounts
• SpaceXSI: Team Bots
• docs.x.ai: Team Bots
• Grok Bot changelog
• docs.x.ai: Approvals, security, and privacy
• docs.x.ai: Grok Bot security
• SpaceXSI: How SpaceXSI is using Grok Bot to scale customer support
• Builder.io: GrokBot is amazing. Here's why I stopped using it.

*Not sure whether an always-on agent belongs in your stack, or which lab's version to trust with your logins? Take the free 2-minute quiz and get matched. Then read the other always-on agent that launched this fall, with its own computer per agent and its own blanks on the price list: OpenAI's Dots.*