Grok Build Data Leak Exposes 35,000 Apps Including SpaceXSI's Confidential Business Plans

Massive Data Leak Affects SpaceXSI's Own Platform

Grok Build, the AI-powered coding agent from SpaceXSI, is at the center of a significant data leak affecting over 35,000 applications. Security research firm Claroc discovered that 35,578 of 37,571 Grok-built apps were accessible to anyone without authentication1

1

2

2

. The exposed applications contained everything from simple games to critical business data, with approximately 2% rated as high or critical severity. Ironically, SpaceXSI—the Elon Musk-owned company behind the platform—had 24 of its own internal apps compromised, revealing confidential onboarding plans, a sales territory map covering 594 accounts, internal system addresses for sales team tools, and advertising proposals marked confidential1

1

.

Critical Design Flaw in URL-Sharing Mechanism

The root cause stems from a fundamental design flaw in how Grok Build handles app sharing. While the platform randomizes URLs using "dice words"—randomly generated phrases like "failing chassis traders"—to prevent easy guessing, this security measure proves ineffective1

1

. Each published app receives its own subdomain of grok.me, and when visitors access an app, their computers perform DNS lookups. Passive DNS services record these lookups, creating searchable lists that anyone can query. Claroc used common subdomain discovery methods to enumerate thousands of live Grok Build projects, with about half following the predictable four-random-word pattern2

2

. This capability-based access control approach—where URL possession serves as authorization—completely fails when the URLs themselves become discoverable through DNS records.

Fortune 500 Companies and Defense Contractors Hit

The confidential data exposure extends far beyond SpaceXSI. Claroc identified sensitive internal data from 30 organizations across construction, energy, healthcare, and defense sectors. Twenty-six of these organizations have more than 1,000 employees, and six are Fortune 500 companies2

2

. Specific examples include a Peruvian construction company's occupational-health dashboard containing 3,784 worker medical records, Tesla's service-center scorecard naming 22 technicians with productivity and bonus status, and a US Navy ship engineering contractor's roster of 108 people on a logistics support contract1

1

. A leading luxury cruise company exposed galley checklists listing crew by full name and employee IDs, while a healthcare company's hospital staff app revealed approximately 3,565 employees by name and ID1

1

.

AI-Generated Apps Security Risks Amplified

Grok Build's rapid app development feature—which can create working applications from simple text descriptions within minutes—accelerates the creation of security vulnerabilities. When Claroc tested the platform with two mock apps containing made-up data, both published with the default "Anyone with the link" access setting, the results were alarming2

2

. A contact list app received no sign-in protection, allowing researchers to add and view data from private browser windows. A customer list app did receive automatically-generated sign-in buttons for Google and X accounts, but the code added every person who signed in to the authorized team before verification, rendering the security check meaningless. The AI-generated apps security risks extend to catastrophic failures like a contractor's timesheet whose code stored every employee's password in plain text2

2

.

SpaceXSI's Response and Ongoing Exposure

After Claroc reported their findings through SpaceXSI's security.txt file on Sunday, the company took down all 24 staff apps within hours of media inquiries but left other users' exposed apps accessible1

1

2

2

. SpaceXSI stated that Grok apps remain private until creators share them and that publishing requires multiple steps including choosing between "Only you" and "Anyone with the link" options, plus accepting a legal disclaimer. The company claimed employee accounts default to "Only you"2

2

. However, the existence of 24 open staff apps contradicts this assertion. Claroc's co-founder Asaf Aprozper began notifying affected companies directly, with the luxury cruise company's CISO immediately assembling their security team to contain the exposure1

1

. Claroc recommended making "Only you" the universal default and adding domain-restricted access options for enterprise users2

2

.