Grok Build apps exposed SpaceXSI and patient data

Claroc says 35,578 apps built with Grok Build opened for anyone without sign-in, including 24 made by SpaceXSI and X staff.
What it means for founders
- Audit what your team has published. A Grok app goes live from the chat window that built it, with no request to IT, Claroc notes. If you have passive DNS access, search it for grok.me and open each hit signed out.
- Treat the access picker as a security control. On the paid account Claroc tested, the publish panel arrived with the link-sharing option preselected. For anything internal, pick Only you or Invited users; Claroc's test app then answered strangers with a 401 error.
- Test the sign-in code Grok writes. In a four-person sales team list Claroc built as a test, Grok wired up Google and X login but enrolled each new account on the team before checking membership, so an uninvited Google user could read, edit and delete records. Test AI-built internal tools from an outside account.
- Watch the default for ordinary users. SpaceXSI defended its publish flow and has not promised a change, Claroc says.
How Grok Build links landed in DNS databases
Every published Grok app gets its own hostname under grok.me, often four random words; about half the names Claroc saw were. When an invited person opens the link, the lookup reaches commercial passive DNS services, which subscribers can search by pattern. Claroc's subdomain discovery listed 43,601 names. Of the 37,571 apps still online, 35,578, or 94.7%, opened without a login.
What the exposed apps held
The staff apps included an onboarding plan naming internal systems and 19 of the company's chat channels, a map of 594 strategic sales accounts with pipeline data, and eight X advertising proposals, one aimed at a federal agency and labeled confidential.
Claroc estimates that about one in ten open apps held sensitive or internal data, and roughly 2% rated high or critical. Its confirmed cases include an ICU patient briefing, a hospital group's staff roster above 3,500 names, a fire department shift board and a timesheet whose script exposed 18 plain-text passwords. IBTimes reports Claroc also found a Peruvian construction firm's dashboard with 3,784 worker medical exams. Claroc counts 30 other affected organizations, six in the Fortune 500, and names none. Per IBTimes, Claroc co-founder Asaf Aprozper granted that Lovable, Bolt and Replit app addresses also turn up through DNS, adding that "Grok is where we found by far the most sensitive company data."
How SpaceXSI responded
Claroc reported the issue on November 8. According to Claroc, SpaceXSI told a reporter that apps stay private until shared, that publishing takes several steps including a legal disclaimer, and that staff accounts start at Only you. After Claroc's report it pulled the 24 staff apps, but as of November 10 not other users' apps. SpaceXSI gave no commitment to reach owners, so Claroc is notifying them itself.
The Decoder notes that Claroc sells security tools for AI-built apps. It also recalls a separate July incident with a different product, the Grok Build terminal tool, which sent user files to company servers. Enki covered a related risk in Claude Code approvals leaking secrets.
What we don't know yet
Claroc limited its edit and delete tests to two apps it built itself, so the number of open apps that also let strangers change data is unknown. SpaceXSI has not said whether it will contact owners or change the default for ordinary accounts.
The story
In a report published November 10, security firm Claroc says 35,578 apps built with Grok Build, the web app builder inside SpaceXSI's Grok assistant, opened for anyone without sign-in. SpaceXSI and X staff built 24 of them.
Sources
Tools in this story
We may earn a commission if you sign up through our links. It never affects our ratings or which stories we cover.
The AI chatbot with live search across X
More in Policy & Safety
- Notion AI chats exposed in Fireworks AI breachHackers who breached Fireworks AI got Notion AI chat transcripts from under 1% of workspaces. What to check and rotate now.TechCrunch · 5d ago

- United bans unauthorized AI agents from bookingUnited will block unauthorized personal AI agents from booking its flights, citing shared passwords and FAA check-in rules.TechCrunch · 5d ago

- New York opens AI lab registration under RAISE Act

For founders: Most startups sit outside it. The heavy duties need frontier scale compute and over $500 million in revenue, and a company calling a model by API is not the…
TechCrunch · 7d ago - eBay appears to block Meta's Muse shopping agent

For founders: The terms reach your users. eBay's terms put the promise on whoever accesses its services, so a customer who points an unapproved agent at eBay may be acting…
The Verge · 7d ago - Nvidia board investigates CEO's ties to alleged spy

For founders: Export policy could tighten. Nvidia's case for selling into China rests partly on its promise to keep chips away from barred buyers.
TechCrunch · 9d ago