A recent investigation by security firm Claroc has revealed that a staggering 35,578 Grok-built applications were accessible online without requiring user authentication. The discovery, announced by researchers yesterday, highlights significant concerns over data privacy and security.
Widespread Data Exposure
Among the exposed applications, 24 were developed by employees of the social media platform X and SpaceXSI, the creators of the Grok AI assistant. Although many of these apps consisted of games, landing pages, and calculators, approximately 10% contained sensitive business or personal information.
Notably, some of these apps included a confidential onboarding plan for a new sales hire, a sales territory map with 594 accounts, and a marketing proposal to a U.S. federal agency labeled confidential. While SpaceXSI has since removed the staff-developed apps, other users’ applications remain accessible.
Impact on Multiple Industries
The investigation identified internal data from 30 other organizations spanning various sectors, including construction, energy, healthcare, and defense. Of these, 26 companies employ over 1,000 people, and six are listed on the Fortune 500. According to the report, approximately 2% of these open apps contained information deemed high or critical in sensitivity.
Examples of exposed data include a hospital group’s employee directory, an intensive care briefing, a fire department’s live shift schedule, and a contractor’s timesheet with plaintext passwords. The ease of creating apps via Grok’s chat-based builder, available to all Grok plan users since August 19, has raised concerns about app security settings.
Security Concerns and Recommendations
Each Grok app is hosted on a unique subdomain of grok.me, with access settings determined by the creator. Despite the option to restrict access, many apps were publicly accessible due to default settings. Researchers employed common subdomain discovery techniques to uncover app names, emphasizing the need for improved security measures.
Claroc recommends organizations identify which AI app builders their employees use, review DNS data for potential exposures, and ensure apps containing internal data have appropriate access controls. Attempting to sign in with external accounts can also help identify security gaps.
Future Outlook and Company Response
The researchers have reported their findings to SpaceXSI and recommended making “Only you” the default access setting. Although SpaceXSI responded by emphasizing that apps remain private until shared, they have not committed to notifying affected app owners. Claroc has begun contacting impacted organizations directly.
This incident underscores the importance of robust security practices in the development and deployment of AI-driven applications. As AI technology continues to evolve, organizations must prioritize data protection to prevent unauthorized access and potential breaches.
