Grok Build Leak Exposed Thousands of AI-Built Apps, Including SpaceXSI's Own Sales Plans and Patient Records, Researchers Say

A popular tool that lets anyone build software simply by describing it to an AI has exposed confidential information from thousands of applications, including internal documents from the company that makes the tool, according to security researchers.

Security firm Claroc found that apps built with Grok Build, an AI "vibe-coding" agent operated by SpaceXSI, could be accessed by anyone who found their links, exposing internal business plans, medical results, employee details and passwords, according to Dataconomy and Cybernews, which reported the findings on Wednesday, November 11. SpaceXSI is the company formed from the merger of SpaceX and xAI, the developer of the Grok AI models.

The scale of the exposure

Claroc reviewed 43,551 web addresses under the grok.me domain and found that 35,578 of 37,571 active Grok-built apps could be opened without signing in, according to Cybernews. About one in ten publicly accessible apps held sensitive personal or internal business data. Estimates put the total number of Grok-built apps at more than 40,000, although no official count has been published.

The problem stems from how apps are shared. Grok Build lets users share applications through links, and while the platform randomises the URLs, a design decision allowed anyone to enumerate thousands of live projects, according to Dataconomy. In most cases, those projects could be opened to view their data.

SpaceXSI's own data

Among the most striking findings was that at least 24 apps came from employees of SpaceXSI and its social media platform X, according to Cybernews. The exposed material included an enterprise sales onboarding plan, an internal map of 594 strategic accounts, eight advertising proposals and several media plans.

One advertising proposal was addressed to a US federal agency and marked confidential. Four live advertiser dashboards showed campaign spending, including progress towards a $1.6 million sales target, and seven campaign reports covered advertisers in Brazil, Japan, South Korea and Thailand. Copies of an internal X brand safety playbook were also exposed.

"We were shocked. The company that makes Grok had its own internal material open to anyone holding the link," the researchers said, according to Cybernews.

Sensitive data from other organisations

Claroc identified 30 other organisations whose information was exposed, according to Cybernews, several of them reportedly Fortune 500 companies with more than 1,000 employees each. Their names were not released.

The exposed material included a hospital group's event app containing a searchable roster of more than 3,500 employees, and an intensive care briefing app that showed one patient's laboratory results, imaging and ventilator status. A cruise ship inspection checklist listed crew members by full name and employee ID, and a fire department's shift board showed firefighters' names, badge numbers and working hours.

A contractor's timesheet app contained plain-text passwords and credentials for 18 employees, including administrators, embedded in its code, according to Cybernews.

SpaceXSI's response

Claroc reported its findings to SpaceXSI on November 8 and 9, according to Cybernews. SpaceXSI removed all 24 apps created by its own employees, but the researchers said it had not taken similar action on apps belonging to other users and organisations.

SpaceXSI said Grok apps stay private unless creators choose to share them, according to Cybernews. Its publishing flow lets creators choose between "Only you" and "Anyone with the link" when they select "Publish."

A pattern across AI app builders

The researchers said the problem was not unique to Grok. "Last month it was Meta's Muse," they said, according to Cybernews, referring to a similar issue with another AI app-building tool. The Cloud Security Alliance described the issue as part of a broader "AI app-builder leak pattern," in which apps created with AI tools are public by default or easily discoverable.

Vibe coding, in which people describe the software they want and AI generates it, has become extremely popular, allowing people without programming skills to build apps for work and personal use. But many of these users do not understand security concepts such as access control, and may put sensitive data into apps without realising who can see them.

The shadow AI problem

The leak highlights the risks of so-called shadow AI, in which employees use AI tools without the knowledge or approval of their organisation's IT and security teams. Employees may build apps to manage tasks, track data or share information with colleagues, unintentionally exposing confidential data.

The researchers recommended that organisations audit which AI app-building platforms their employees use and add safeguards against data leaks from AI-built apps, according to Cybernews.

What it means for Indian businesses

Indian companies and employees have rapidly adopted AI tools, including app builders and coding assistants. The Grok Build leak is a warning that sensitive business and personal data can be exposed through these tools if they are not used carefully. India's Digital Personal Data Protection Act places obligations on organisations to protect personal data, and a leak of personal data through an employee-built app could expose companies to regulatory penalties.

For Indian IT services companies, which build software for global clients, the incident also highlights the importance of secure development practices as AI-generated code becomes more common.

Lessons for users

Users of AI app builders should check sharing settings before publishing apps, avoid putting sensitive data such as passwords, medical records or confidential documents into apps unless they are certain of who can access them, and remove apps they no longer need. Platforms, for their part, may need to adopt safer defaults and stronger protections against link enumeration.

Why links are not locks

Sharing content through unguessable links is a common convenience on the internet, used by document-sharing, file storage and design tools. But security experts have long warned that a link is not a lock: if links can be guessed, enumerated or forwarded, anyone who obtains them can see the content.

For applications that hold sensitive data, proper authentication, requiring users to sign in and checking whether they are authorised, is essential. The Grok Build case suggests that many users of AI app builders did not apply such protections, possibly because they assumed their apps were private.

What comes next

Pressure is likely to grow on SpaceXSI to address the exposure of other users' apps and to change how Grok Build handles sharing. Regulators may also take an interest, particularly where personal and medical data were exposed. More broadly, the incident adds to concerns about the security of AI-generated software, as vibe coding moves from a novelty to a widely used tool in businesses around the world.