Insights AI News Grok Build Shared Link Data Leak: How to Protect Your Data
Grok Build Shared Link Data Leak: How to Protect Your Data (AI-generated)AI-generated

AI News

19 Nov 2026

Read 12 min

Grok Build Shared Link Data Leak: How to Protect Your Data

Grok Build shared link data leak exposes secrets; restrict access to protect sensitive apps and data.

The Grok Build shared link data leak shows why a hard-to-guess web address is not the same as private access. According to research from security firm Claroc, thousands of apps exposed business records and personal data. Users should restrict access to invited people and review what their published apps contain.

Key takeaways

  • Claroc found 43,601 addresses ending in grok.me and reported that more than 35,000 apps were publicly accessible.
  • Exposed records included worker medical exams, employee details, sales plans, and confidential business proposals.
  • Random app names can appear in passive DNS records, making supposedly secret addresses discoverable.
  • Claroc and a company representative gave conflicting accounts of Grok Build’s default sharing settings.

What caused the Grok Build shared link data leak?

The reported exposure came from discoverable app addresses combined with access settings that let anyone holding a link open the app.

Grok Build lets users create working apps through text or speech. Sharing links makes it easy to send those apps to colleagues or testers. But when an app contains private records, that convenience can expose more than its creator expects.

Claroc found sensitive information in apps belonging to businesses, including Fortune 500 companies and a defense contractor. It also identified 24 apps containing internal material from SpaceXSI, described in the source report as the merged SpaceX and xAI.

Claroc notified SpaceXSI on Sunday before the report appeared. After receiving no response, a Claroc official tagged Elon Musk and Grok Manager Matt Palmer on X. According to Ars Technica, the 24 SpaceXSI apps became unavailable after Ars sent questions. Almost all other exposed data remained accessible when the report was published.

What information was exposed?

Claroc reported finding medical records, employee rosters, financial details, and confidential company documents inside accessible apps.

The SpaceXSI material included a confidential enterprise-sales onboarding plan, a territory map covering 594 accounts, internal addresses for sales tools, and advertising proposals. Other findings show how widely the exposure reached:

  • A Peruvian construction company had an occupational-health dashboard containing 3,784 worker medical exams.
  • A Tesla service-center scorecard named 22 technicians and listed productivity and bonus status.
  • A US-Navy ship engineering contractor had a roster of 108 people on a Navy logistics support contract.
  • A healthcare company had an app listing about 3,565 employees by name and ID.
  • A US home construction company had a sales callsheet covering about 583 prospects, including budgets and financing notes.
  • An auto transport company had about 4,000 orders showing prices and margins.

Claroc also found cruise-line crew details, confidential customer proposals, and an explosives-building loading procedure. These records concern people who may never have used Grok Build themselves.

Asaf Aprozper, Claroc’s co-founder and CTO, said he was notifying affected companies. He said the luxury cruise company involved brought in its security team after receiving the findings.

Why did random app addresses fail to keep data private?

The random words appeared in the hostname, where passive DNS services could record them, rather than in a URL path that DNS does not see.

A hostname is the part of an address that identifies the site. Grok Build places each app’s random words before grok.me. DNS translates that hostname into the network address needed to reach the app.

Passive DNS services collect observations of DNS activity. Their searchable records can therefore reveal app hostnames without anyone guessing the random words. They do not necessarily capture every hostname, but Claroc found thousands through this route.

This matters because link-based access treats possession of the address as permission. Aprozper said a search returned 43,601 grok.me addresses, with more than 35,000 apps publicly accessible.

By contrast, Google Docs sharing links put the hard-to-guess identifier after the domain, in the path. DNS sees the domain but not that identifier. Moving a secret into the path prevents this particular discovery method; it does not stop recipients from forwarding the link.

The Grok Build shared link data leak therefore reflects an access-design problem, not simply weak random words.

Are Grok Build apps private by default?

The default setting remains disputed: Claroc reported public link access during its tests, while a company representative said apps start private.

A representative with Cursor, which the source says SpaceXSI recently acquired, told Ars that Grok Build apps are private by default and become public only when creators choose to share them.

Aprozper reported a different result. His tests over the preceding week showed “Anyone with the link” selected by default, requiring users to switch to “invited users” or “only you.”

The representative did not answer follow-up questions described in the report. Users should inspect each app’s actual access setting rather than rely on either account of the default.

How can you protect your Grok Build data?

Restrict access first, then review the app’s contents and decide whether the exposure requires a wider security response.

Check every published app

For personal work, choose “only you.” For team work, use the invited-user option and grant access only to people who need it. Reserve “Anyone with the link” for material you genuinely intend to make broadly accessible.

Review older apps as well as current projects. A forgotten test app may still contain real records copied in during development.

Remove sensitive records from public projects

Look for employee names, IDs, medical information, customer budgets, pricing, internal tool addresses, and confidential proposals. These are among the data types Claroc reported finding.

Use sample data when demonstrating an app publicly. A working interface rarely needs real employee or customer records to show what it does.

Escalate possible exposure

If company information was accessible, involve your security or IT team. Record which apps and data were involved so the team can assess the next steps. If you discover exposed credentials, treat them as potentially compromised and replace them.

Changing a setting blocks future access under that setting; it cannot undo earlier viewing or copying. The report does not establish who else accessed the exposed apps.

Responding to the Grok Build shared link data leak should therefore involve more than renaming an app or creating another random address.

Does the same risk affect other AI app builders?

Other builders can face the same hostname-discovery risk, although discoverability alone does not prove that an app exposes private data.

Aprozper said Lovable, Bolt, and Replit also give apps individual web addresses that can appear in DNS records. He said Grok contained by far the most sensitive company information his research found.

The source contrasts this approach with Claude’s published artifacts, ChatGPT’s shared conversations, and Gemini’s shared links. Those place their random identifiers in URL paths. When comparing AI coding chatbots, assess publishing controls alongside coding features.

Our take

This report shows that fast app creation needs equally clear controls over who can read the result.

Compared with the earlier Muse link-sharing incident mentioned in the source, the striking finding here is the reported exposure of the platform provider’s own internal documents. It suggests that even people close to an AI product can misunderstand its publishing boundaries.

Companies and developers need to review shared apps, while employees and consumers may be affected without having created one. Readers in Germany and the EU should apply the same checks to business and personal records; the report does not establish a separate German impact.

One open question is whether some published apps contained demonstration records rather than live data. That could change individual impact assessments, but it would not resolve the discovery weakness. In the coming weeks, watch for clearer defaults, customer notifications, and evidence that containment extends beyond SpaceXSI’s own apps.

The Grok Build shared link data leak is a reminder to treat secret links as shareable access, not proof of privacy. Check permissions, remove sensitive records from public apps, and involve your security team if real business or personal data was exposed.

(Source: https://arstechnica.com/security/2026/11/grok-build-coding-agent-leaks-oodles-of-confidential-info-some-belonging-to-spacexsi/)

For more news: Click Here

FAQ

Q: What caused the Grok Build shared link data leak? A: Grok Build app addresses used random words in the hostname, where passive DNS services could record and expose them. Apps set to allow anyone with the link could then be opened by people who found those addresses. Q: What kinds of information were exposed? A: Claroc reported finding medical exams, employee names and ID numbers, sales records, customer budgets, and confidential business proposals. Some apps also contained internal SpaceXSI material, including sales plans and advertising proposals. Q: How many Grok Build links did Claroc find? A: Claroc said a passive DNS search returned 43,601 addresses ending in grok.me. It reported that more than 35,000 apps were publicly accessible. Q: Why didn’t random words in the app address keep it private? A: The random words appeared before grok.me, in the hostname that DNS services process and may record. Putting a secret identifier in a URL path, as the article says Google Docs does, avoids this particular discovery method. Q: Are Grok Build apps private by default? A: The accounts conflict: a Cursor representative said apps are private by default, while Claroc reported that “Anyone with the link” was selected by default in its tests. The article recommends checking each app’s actual access setting. Q: How can users protect data in Grok Build apps? A: Set personal apps to “only you” and team apps to “invited users,” reserving “Anyone with the link” for content intended to be broadly accessible. Review older projects and remove real employee, customer, medical, or confidential business data from public apps. Q: What should a company do if its information was exposed in the Grok Build shared link data leak? A: Involve the security or IT team and document which apps and information may have been accessible. If credentials were exposed, treat them as potentially compromised and replace them. Q: Can changing an app’s access setting undo an earlier exposure? A: Changing the setting can block future access under that setting, but it cannot undo earlier viewing or copying. The report does not establish who accessed or copied the exposed apps.

Keep reading, keep building

Three things from our own workshop that fit this topic:

Our own products. Tech Moments is our newsletter on AI and tech in plain language, one email, no noise. Subscribe below.

Contents