A business app can look ready for work while leaving its data open to strangers.
A security investigation by Claroc found that thousands of applications created with Grok Build were accessible without authentication, exposing sensitive company information. Researchers also identified 24 publicly accessible applications created by SpaceXSI and X staff.
SpaceXSI said employee apps default to private and publishing requires deliberate steps, according to the response reproduced by Claroc. The researchers said the company subsequently removed the 24 staff-built apps.
The findings also raise wider concerns about the security of AI-generated software as businesses increasingly turn to coding agents and app builders to create business tools. Making software easier to build does not necessarily make it secure, particularly when applications handle sensitive business information without undergoing proper security reviews.
What is Grok Build
Before getting into the security problems, it’s worth clearing up one thing: Grok and Grok Build aren’t the same product.
Grok is SpaceXSI’s AI assistant, which users can turn to for answers, research, content generation, and other tasks, although it can also generate code. Grok Build, on the other hand, is designed to help users create software by describing what they want in ordinary language. Think of it as an AI-powered app builder rather than a chatbot.
Here’s how they differ:
| Grok | Grok Build | |
|---|---|---|
| What it does | Answers questions and performs AI-assisted tasks | Builds software from user instructions |
| How users interact with it | Through conversations | Through conversations. |
| What it produces | Text, images, analysis, code, and other content | Working applications, websites, and dashboards |
With Grok Build, users don’t necessarily need programming experience to create something functional. They can describe an application, let the AI generate it, and publish it online, where it receives its own grok.me address.
That last part is particularly important to this story.
How Grok Build spilled the tea on users
The exposure stemmed from two problems: Grok Build’s default sharing settings and the way it generated application URLs.
When publishing an application, users could choose who could access it. In Claroc’s test account, “Anyone with the link” was preselected, allowing visitors to open the published app without signing in. Users who left this setting unchanged could unknowingly expose sensitive business information.
Outsiders did not need to guess the links. Grok proposed application addresses made from four random words, although not all published apps used that format. The names appeared in app subdomains discoverable through passive DNS records.
Passive DNS services collect records of domain-name lookups, allowing researchers to find previously accessed domains. Claroc used these records to identify application addresses without needing their owners to share them.
Claroc identified 43,601 application addresses, of which 37,571 were live and 35,578 opened without authentication. Most accessible apps were games, landing pages, or calculators, but some exposed employee records, patient information, confidential business documents, and plaintext passwords.
The 24 applications built by SpaceXSI and X staff exposed information including sales territory maps, employee onboarding plans, and confidential advertising proposals.
Grok Build isn’t even alone in this
Other AI platforms have also faced problems with shared content becoming accessible beyond its intended audience. In 2025, OpenAI removed a feature that allowed shared ChatGPT conversations to appear in Google search results. Anthropic faced similar scrutiny in July after shared Claude conversations and Artifacts containing sensitive information appeared in search results.
A closer, and even more recent, comparison is Meta’s Muse AI agent. In October, Claroc discovered more than 6,800 shared Muse links, some exposing confidential business documents and personal information. Like Grok Build, Muse’s URL structure made shared content discoverable through DNS records.
Although the incidents differ technically, they highlight a common concern: AI platforms are making content easier to create and share. Still, their sharing mechanisms can expose information users never intended to make widely accessible.
More must-read AI coverage
- SS&C Intralinks DealCentre AI vs. Datasite: Which platform is built for the future of dealmaking?
- SS&C Intralinks FundCentre AI vs. Juniper Square: Which platform better supports modern private markets fund managers?
- Why Data, Not Models, Determines AI Success
- The Rise of the AI-Native Factory: How Physical AI Is Transforming Manufacturing
Why enterprises must care and what they can do
The push to build and deploy software faster is creating security challenges businesses cannot ignore. AI coding agents and app builders allow even non-developers to create tools for everyday operations. But as Grok Build demonstrates, getting an application running quickly doesn’t mean it’s secure.
Small businesses and teams may be particularly exposed. AI app builders appeal to organizations that cannot afford dedicated development teams or expensive software solutions. They can quickly build tools for managing customers, employees, and daily operations. However, limited resources can also mean fewer people available to review these applications before deployment.
Saving development time means little if the resulting applications expose the information they’re supposed to protect.
- Review sharing settings and restrict access to authorized users.
- Test authentication and authorization before deployment.
- Keep passwords and API keys out of application or production code.
- Maintain an inventory of AI-generated applications.
- Require additional security reviews for tools handling customer, employee, and financial data.
AI can speed up software development, but businesses remain responsible for what they deploy. Saving development time means little if the resulting applications expose the information they’re supposed to protect.
Businesses should treat AI-generated applications like any other software handling sensitive information. A working interface is only one part of deployment readiness; teams also need to establish who can access the app and what those users can do.
Start with a short review:
- Assign an owner. Record the app’s purpose, the information it handles, and who approves changes.
- Check access while signed out. Open the published app in a private browser window and confirm that internal information is unavailable.
- Test with an unauthorized account. A successful sign-in should not automatically grant access to company records or administrative functions.
- Review browser-delivered code. Check scripts and configuration for exposed passwords, API keys, and other credentials.
- Repeat checks after changes. New features, integrations, and sharing settings can alter an app’s exposure.
Authentication and authorization need separate checks. A login screen establishes identity; permission checks determine which records and actions that identity can access.
If an app exposes sensitive information, restrict access and involve the security team to assess what was accessible and whether exposed credentials need replacing. Assign responsibility for resolving the findings before the app returns to use.
AI can shorten development time, but deployment approval should depend on tested access controls, an accountable owner, and a process for reviewing changes.
Read more: As AI makes software easier to create and share, Meta’s Pocket app shows how vibe coding is moving into everyday social experiences.