November 11, 2026 – Security researchers have discovered that more than 35,000 applications created with Grok’s AI app-building tool could be opened without a login. Some of these apps contained sensitive business documents, employee information and personal records that were not intended for public access. The findings were published by cybersecurity company Claroc Research in its November 10 security investigation.
Grok Build allows people to create websites, dashboards, games and other applications by describing what they want in everyday language. Users can then publish their projects online and share them with others. According to the company’s official Grok Build announcement, published apps receive individual web addresses under the grok.me domain, with options to control who can open them.
Claroc researchers tested 43,551 Grok app addresses and found that 37,571 were still online. Of those active apps, 35,578, or approximately 94.7%, opened without requiring visitors to sign in. However, this does not mean every accessible app exposed private information. Most were ordinary projects such as games, calculators and landing pages.
The investigation found a smaller group of applications containing potentially sensitive information. Researchers identified business records, employee lists, customer information and healthcare-related data. Claroc estimated that approximately 650 to 800 apps contained private data, although this estimate has not been independently verified.
Among the findings were 24 publicly accessible apps created by employees of SpaceXSI and X. These included internal sales materials, an employee onboarding document and advertising proposals. Researchers also identified confidential information associated with 30 other organizations, including large businesses operating in technology, healthcare, transportation and other industries.
The problem was connected to how Grok Build creates and shares web addresses. Although its app links can contain random words that make them difficult to guess, the names can still appear in searchable internet address records known as DNS data. Researchers used these records to discover published apps without receiving their links directly from their creators.
Claroc also created two test applications to examine the security settings. One app stored information without requiring a login. Another included Google sign-in, but the researchers found that an unrelated Google account could still access and change its records. These tests were performed on applications created by Claroc itself, not on other people’s projects. The findings demonstrate why checking a user’s identity is different from confirming that the user has permission to access specific information. TechBonafide’s guide to API security and access permissions explains this important distinction.
The company behind Grok has challenged parts of the findings. According to reporting by Ars Technica, a company representative said apps remain private until their creators choose to publish them. Claroc agreed that publishing requires user action but reported that, in its test account, the option allowing anyone with the link to open the app was already selected during publishing. The difference between these statements concerns what happens before publication and which access option is selected when an app is shared.
Claroc first notified the company about the exposure on November 8 and provided further information on November 9. Following the investigation, the 24 identified apps belonging to SpaceXSI and X employees were removed. Ars Technica independently reported that these app addresses no longer worked when its article was published on November 10. However, Claroc said other users’ accessible apps had not all been removed at that time.
Several questions remain unanswered. The investigation does not establish that attackers downloaded the exposed information or used it for criminal activity. It is also unclear how many outside visitors accessed the apps before the findings became public, whether every affected organization has been notified, or how many of the identified apps remain accessible. The research shows a serious risk of unintended access, but it should not be treated as proof that all 35,578 apps suffered a malicious data breach.
For developers, businesses and people using AI tools to create applications, the findings highlight the importance of checking privacy settings before publishing. Grok Build users handling private information should review existing projects, choose restricted access options such as Only you or Invited users, and confirm that unauthorized accounts cannot open or change stored data. Applications containing passwords, customer records or confidential documents should also receive a security review before being shared.
The investigation highlights a growing challenge as AI makes software development easier for people without traditional programming experience. Creating an application in minutes does not automatically make it secure. Clear access permissions, careful handling of sensitive information and regular security checks remain essential, regardless of how an application was built.
