Grok Build Leak Exposes 35,000 Apps — Including SpaceXSI's Own Confidential Sales Plans

More than 35,000 apps built with Grok Build — the AI app builder from SpaceXSI, Elon Musk's merged SpaceX and xAI — were sitting open on the public internet, readable by anyone who found their web address. No password. No sign-in. Among them were 24 apps built by SpaceXSI's own employees, holding confidential sales plans, internal system addresses, and an advertising proposal for a US federal agency. Security firm Claroc published the findings on Tuesday, November 10, 2026, and Ars Technica reported them the same day.

The Short Answer

Security firm Claroc found 35,578 apps built with Grok Build publicly accessible without any sign-in — 94.7 percent of the 37,571 apps still online that it checked. The exposed apps included 24 built by SpaceXSI and X staff, containing a confidential onboarding plan for a new enterprise sales hire, a sales territory map covering 594 customer accounts, and advertising proposals. Other exposed apps held medical exam results, employee names and ID numbers, business plans, and plaintext passwords. SpaceXSI removed its own 24 apps within hours of Ars Technica asking questions, but apps belonging to other users remained open.

How 35,000 Apps Ended Up Exposed

Grok Build is a "vibe-coding" agent: a user describes the app they want in a chat, Grok writes the code, and the app can be published on the spot. Each published app gets its own hostname under grok.me, usually four random words in front of the domain. The design assumes those addresses are secret — only people with the exact link can open the app.

That assumption failed. Claroc surveyed the grok.me namespace and listed 43,601 app addresses. Of the 37,571 still online when it tried them, 35,578 opened for anyone, with no sign-in. Only 1,993 returned an access-denied error. The cause is that the random part of the address sits in the hostname, where public DNS records can reveal it — so the addresses can be enumerated. Claroc calls this class of leak "GhostShares," and says it found the same weakness last month in Meta's Muse.

Claroc estimates that about 10 percent of the exposed apps contained internal business data or personal information, and about 2 percent were rated highly sensitive. The sampling found corporate business plans, advertising proposals, employee names and identification numbers, medical examination records and patient lab values, staff rosters and shift schedules, galley checklists from a celebrity cruise line, and plaintext passwords.

SpaceXSI's Own Apps Were Among the Exposed

The most embarrassing detail is that the company that makes Grok exposed itself. Claroc found 24 open apps built by SpaceXSI and X employees, including a confidential onboarding plan for a new enterprise sales hire, a sales territory map of 594 accounts, internal network addresses for sales team utilities, and an X advertising proposal for a US federal agency marked confidential. Claroc's co-founder and chief technology officer, Asaf Aprozper, told Ars Technica the firm is working to notify the affected companies.

Claroc first reported the problem to SpaceXSI through the address in the company's security.txt file on November 8, then sent details of the other staff apps on November 9. SpaceXSI never replied, and all 24 of its apps were still exposed on Monday evening. Within a few hours of Ars Technica emailing questions, SpaceXSI began pulling them down. At publication, the links for all 24 returned "app not found" — but the apps belonging to other Grok users and organizations stayed online, and SpaceXSI has not said how it will handle them or notify affected organizations.

SpaceXSI responded to the disclosure by saying that apps are private until their creator chooses to share them, that publishing takes multiple steps including a legal notice, and that employee accounts default to "Only you" — yet its own staff left 24 apps open.

A Pattern, Not a One-Off

This is at least the third privacy incident for SpaceXSI involving Grok. In August 2025, Forbes found more than 370,000 shared Grok conversations in Google Search. In July 2026, a researcher named Cereblab found that the Grok Build command-line tool packaged users' whole code repositories — including Git history and committed secrets — and uploaded them to company cloud storage. The new finding shows the problem is structural: unlisted links are not access control when the addresses can be enumerated.

The lesson applies well beyond Grok. Vibe-coding tools from every major lab now let anyone publish an app in minutes, and the defaults decide whether that app is private or public. Microsoft's PDD Kit takes the opposite approach — formally verifying AI-written code before it ships — while researchers at MIT built WorldParticle to simulate the physical world AI systems increasingly act in. As AI-generated apps spread, the gap between "shared with a link" and "public to everyone" is becoming one of the industry's most common security holes. You can follow the latest AI security and Grok news as Claroc notifies affected organizations.

Read Also

More Tech & AI coverage from TrendWire Daily: Microsoft's PDD Kit Uses Rust to Verify AI-Written Code and Anthropic Launches Claude Fable 5.2 and Mythos 5.2.

Frequently Asked Questions

What is Grok Build?

Grok Build is a "vibe-coding" agent from SpaceXSI — the company formed by the merger of SpaceX and xAI, which Musk renamed in October 2026. Users describe an app in a chat, Grok writes the code, and the app can be published to its own address under grok.me.

How many Grok Build apps were exposed?

Claroc listed 43,601 app addresses under grok.me. Of the 37,571 still online, 35,578 — 94.7 percent — opened for anyone without sign-in. Only 1,993 returned an access-denied error.

What kind of data was exposed?

Business plans, advertising proposals, employee names and ID numbers, medical exam results and patient lab values, staff rosters, shift schedules, and plaintext passwords. Claroc estimates about 10 percent of the exposed apps held internal business data or personal information.

Did SpaceXSI's own data leak?

Yes. Twenty-four of the exposed apps were built by SpaceXSI and X staff. They included a confidential onboarding plan for a new enterprise sales hire, a sales territory map of 594 customer accounts, internal system addresses, and an advertising proposal for a US federal agency marked confidential.

Has SpaceXSI fixed the problem?

SpaceXSI removed its own 24 apps within hours of Ars Technica asking questions, but apps belonging to other users and organizations remained open at publication. The company has not disclosed how it will handle those apps or notify affected organizations.

How can Grok Build users protect their apps?

Check your published apps' sharing settings and confirm each one is set to private unless you intend it to be public. Remember that an unlisted link is not real access control — if the address can be discovered, anyone can open the app.

Related: Microsoft's PDD Kit Verifies AI-Written Code | MIT WorldParticle Simulates Sand, Cloth and Proteins | Anthropic Launches Claude Fable 5.2 and Mythos 5.2