Grok AI App Builder Exposes Over 35,000 Apps, Leaking Confidential Business Data and Passwords
Apps built with Grok AI App Builder are publicly accessible, exposing sensitive business data, employee details, and plain-text passwords for over 35,000 organizations.
Thousands of applications created using Grok AI's App Builder have been found to be publicly accessible online, inadvertently leaking confidential business data, staff details, plain-text passwords, and sensitive patient information. Researchers from Claroc discovered that no sophisticated hacking techniques were required to access this data; the apps became publicly searchable through passive DNS records after their initial use.
Grok's App Builder allows users to create and publish chat-generated applications under the grok.me domain. The parent company, SpaceXSI (formerly xAI), was contacted for comment. Claroc researchers analyzed a list of over 43,500 grok.me addresses on November 8, 2026. By simply opening the first page of each app in a private browser window, they found that a staggering 35,578 out of 37,571 apps (94.7%) were accessible without any login requirements, while only 1,993 prompted for credentials.
The passive DNS records, which log website name lookups, provided the list of app addresses. Each Grok app receives a unique, albeit hard-to-guess, four-random-word name that enters these records once the app is accessed. This same method was previously used to expose data from Meta's Muse AI. The exposed apps spanned various categories, including games, landing pages, and calculators, with approximately 10% containing sensitive or internal data, and about 2% deemed high or critical risk.
Specific examples of exposed data included an intensive care briefing with patient lab results and ventilator status, a hospital group's employee roster of over 3,500 individuals, and a fire department's shift board containing badge numbers. One contractor's timesheet app contained 18 staff accounts with plain-text passwords embedded in the browser-downloaded code. SpaceXSI and xAI also had 24 of their own apps exposed, including a confidential onboarding plan and a sales map of 594 accounts.
Of the organizations affected, 30 identifiable entities had internal data exposed. Notably, 26 of these have over 1,000 employees, and six are Fortune 500 companies. While SpaceXSI took down its own exposed apps following the report, it did not address the user-created applications. Claroc noted that paid-account apps defaulted to public link access, allowing anyone to view and modify records without proper authentication.
Even apps with Google sign-in checks were found to be vulnerable, as they did not verify if the user belonged to the intended team. This meant a stranger with a Google account could potentially read, edit, or delete customer records. SpaceXSI stated that apps remain private until explicitly shared and that publishing involves several steps, a claim Claroc refuted by pointing out that all owners of the exposed apps had followed the same publishing process.
This incident is not the first time Grok has faced data exposure issues. In 2025, over 370,000 shared Grok chats were found to be publicly indexed by Google. The researchers recommend that teams identify all AI app builders used by their staff, as tools like Lovable, Replit, and Grok often bypass IT approval. They also advise checking app access settings, testing sign-in procedures with external accounts, and regularly reviewing scripts for exposed passwords, urging users to unpublish any unused or unowned applications.
The full research findings are available on the Claroc blog. This incident highlights a growing concern around the security of AI-generated applications and the potential for sensitive data leakage when proper access controls and privacy settings are not rigorously implemented and verified.
The new report from Claroc details that 35,578 of the 37,571 Grok-built applications discovered online were publicly accessible without any authentication. Among these, 24 applications developed by staff at X and SpaceX were found to contain sensitive internal business data, including an onboarding plan and a confidential advertising proposal. While SpaceX has since removed its staff-developed applications, the researchers are now directly contacting other affected organizations to notify them of the exposed data, which includes internal business information from at least 30 other companies across various sectors.