Grok Build apps mass exposure: SpaceXSI/xAI Grok — 35,578 published apps open to anyone, some leaking internal data and plaintext passwords (no CVE)
Severity: high · Patch status: partial · Since 2026-11-11
Researchers at Claroc found 37,571 apps built with Grok's chat-based app builder, of which 35,578 were publicly accessible without sign-in. About 1 in 10 contained internal business or personal data, including X/SpaceXSI staff apps (a confidential onboarding plan, a 594-account sales territory map, a federal agency ad proposal), a hospital's 3,500-employee list, an ICU patient briefing, and a contractor's timesheet with plaintext passwords. Root cause: 'Anyone with the link' is pre-selected at publish, and app subdomains are discoverable via passive DNS. Test apps also showed broken auth logic (users added before membership check). SpaceXSI removed the 24 staff apps but did not reply to the report; other apps remain exposed. Related Muse (Meta) passive-DNS sharing issue was fixed in October.
What to do
- Inventory which AI app builders (Grok, Lovable, Replit, v0, Bolt, Base44, Google AI Studio) staff use
- Query passive DNS for builder domains and open results in a private browser to find exposed apps
- Set any Grok app holding internal data to 'Invited users' or 'Only you' and verify strangers get an error page
- Attempt sign-in with an outside account; if it succeeds, add an explicit allow-list
- Inspect app scripts for names, emails and passwords; unpublish ownerless apps
Sources
All live incidents on Threat Watch
