Play store
A security vulnerability in Grok Build, an AI-powered application development tool owned by Elon Musk’s company SpaceXSI, has exposed sensitive information belonging to thousands of users, including hospitals, fire departments and the company’s own internal operations.
Cybersecurity research firm Claroc identified the flaw, which allowed publicly accessible applications and webpages to reveal confidential information without requiring users to have direct links.
Researchers have named the exposed pages “Ghostshares”, referring to applications that could be discovered and accessed by anyone on the internet.
How the vulnerability occurred
Grok Build allows users to create applications by giving instructions to artificial intelligence, a process commonly known as “vibe coding”.
The tool generates application links using random four-word phrases. However, researchers found that it placed these random identifiers before the main domain as subdomains rather than incorporating them into the domain’s path.
According to Claroc, this approach allowed passive DNS servers to record the links, making them searchable through publicly accessible databases and exposing applications that users might have expected to remain private.
Other platforms, including Google’s Gemini, use different URL structures that place random identifiers after the main domain, helping prevent this type of exposure.
Extent of the data exposure
After examining more than 35,000 publicly accessible Grok applications, Claroc researchers found that around 10 per cent contained sensitive or business-related information, while at least 2 per cent included highly sensitive data.
The exposed SpaceXSI materials included employee onboarding plans, sales maps covering 594 strategic accounts, advertising proposals for X (formerly Twitter), live account dashboards and brand safety guidelines.
Researchers also discovered pages containing employee lists from a hospital group with more than 3,500 staff members, fire department shift schedules and employee usernames and passwords.
The findings raised concerns about the privacy and security of organisations using AI-powered development tools to create applications and share information.
Company response
Claroc said it notified SpaceXSI about the vulnerability but initially received no response.
The company later removed 24 of its own exposed internal applications after Elon Musk and relevant officials were tagged in posts on X, according to the researchers.
However, Claroc said there had been no coordinated effort to address unsecured applications belonging to other users and organisations.
The findings highlight the potential security risks of AI-powered application development when privacy settings, access controls and data protection measures are not properly configured.