Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

Grok Build leaks thousands of users' confidential data via DNS-discoverable URLs

43,601 grok.me links found via passive DNS; nearly 95% publicly accessible.

Security firm Claroc found that Grok Build, SpaceXSI's AI-powered coding agent, has been leaking proprietary business information and other sensitive data from thousands of users — including SpaceXSI itself1.

The 2 articles this piece was written from — of 3 independent outlets covering the story.

A passive DNS search returned 43,601 links ending in grok.me, and more than 35,000 of them — nearly 95 percent — were publicly accessible2. Grok Build lets users create web applications by describing what they want in a chat interface and publish them directly, each receiving a unique web address in the format .grok.me with four randomized words. By default, Grok sets link access to "Anyone with the link," though users can change settings to "invited users" or "only you".

How the URLs leak

The randomized addresses were intended to prevent guessing, but the grok.me subdomains can be found through passive DNS services, which collect and index domain names observed in DNS traffic. A design decision by Grok Build's developers makes it possible for anyone to enumerate thousands of live projects, and clicking those links in most cases opens the app and reveals all associated data.

ANALYSIS The combination of a permissive default sharing setting and DNS-discoverable subdomains effectively negates the randomization meant to protect user projects.

Other AI-based app builders, including Lovable Labs Inc., Bolt.new, and Replit AI, give each app its own web address, meaning the same passive-DNS technique could expose them as well. By contrast, Claude's published artifacts, ChatGPT's shared conversations, and Gemini use path-based secret randomness that DNS never sees.

What the data contained

Claroc found exposed data including internal business plans, ad proposals, medical exam results, galley checklists from a celebrity cruise line, and employee names and ID numbers.

SpaceXSI's own apps were not spared. Claroc identified 24 SpaceXSI apps exposing sensitive internal material. A sampling of those documents included an onboarding plan marked confidential for a new enterprise sales team member, a sales territory map of 594 accounts, internal system addresses for sales team tools, and advertising proposals.

Claroc built two test apps to assess whether data in Grok Build apps could be altered. One test app allowed an unauthenticated visitor to modify a shared contact database, and another automatically granted access to anyone signing in with a Google account, even if they were not part of the team.

SpaceXSI's response

Claroc notified SpaceXSI of the privacy spill on Sunday. SpaceXSI began removing the links within a few hours of Ars Technica sending an email with questions. SpaceXSI remediated the issue after disclosure.

A representative said Grok Build apps are private by default and nothing goes public unless the creator chooses to share it. The representative also noted that ChatGPT, Lovable Labs, and Bolt.new let users generate shareable links as well.

Was this useful?

The day’s vector, in your inbox.

Vector Wire Daily: one dispatch each publishing day.