Grok Build, the vibe-coding agent from SpaceXSI — the merged SpaceX and xAI — left roughly 35,000 apps built by its users open to anyone on the internet, according to research from security firm Claroc reported by Ars Technica on November 10. Among the exposed apps were 24 built by SpaceXSI itself, holding its own confidential sales material.
Claroc co-founder and chief technology officer Asaf Aprozper said a passive DNS search returned 43,601 links ending in grok.me, where Grok Build publishes user apps. Of the 37,571 links still online, 35,578 — about 95 percent — opened the corresponding app and all the data stored in it, with no sign-in required. Claroc estimated about one in 10 apps held sensitive business or personal data, roughly 650 to 800 held private data, and about 2 percent held material serious enough to rate high or critical.
SpaceX's own files were among the exposed apps
The 24 apps built by SpaceXSI and X staff included a confidential onboarding plan for a new member of the enterprise sales team, listing internal system addresses, sales tools and 19 internal chat channels. They also included a sales territory map covering 594 strategic accounts and their pipeline, eight X advertising proposals and media plans — one of them to a US federal agency and marked confidential — four live advertiser account boards with campaign spend, and two copies of an X brand safety playbook, plus campaign reports for advertisers in Brazil, Japan, Korea and Thailand.
"The company that makes Grok had its own internal material open to anyone holding the link, and we had never been given it," Aprozper told Ars Technica.
Other people's data was exposed too
The exposure went well beyond SpaceXSI. Claroc's sampling found a hospital staff app listing roughly 3,565 employees by name and ID, a Peruvian construction firm's occupational-health dashboard holding 3,784 worker medical exams, and a Tesla service-center scorecard naming 22 technicians with their productivity and bonus status. There was also a roster of 108 people on a Navy logistics support contract, a galley checklist from a celebrity cruise line, and a load board for an explosives building at a US plant run by a Norwegian-Finnish defense group.
Six affected companies were in the Fortune 500, 26 had more than 1,000 employees, and a defense contractor was among those exposed. Claroc said it was notifying the affected companies directly.
A design flaw, not a broken password
The problem is how Grok Build builds its share links. Each app gets a URL of randomly generated "dice words" placed in the hostname, left of grok.me — something like cinder-crane-sand-frost.grok.me. Hostnames show up in passive DNS databases, which anyone can search, so the supposedly unguessable links could simply be listed and opened. By contrast, Google Docs links, Claude artifacts, shared ChatGPT conversations and Gemini shares keep the secret part in the URL path, which DNS never sees.
A representative for Cursor, the AI coding service SpaceXSI recently acquired, told Ars Technica that Grok Build apps "are private by default and nothing goes public unless the creator chooses to share it." Claroc notified SpaceXSI via its security.txt address on November 8 and sent details of the other staff apps on November 9, but got no response until it tagged Elon Musk and Grok manager Matt Palmer on X. The 24 apps were still exposing data on the evening of November 9; SpaceXSI began removing the links within hours of Ars Technica sending questions, and by publication all 24 returned an "app not found" error. Nearly all of the other users' exposed data, however, remained available.



