Exposed Grok AI apps reveal hidden security risk for businesses
A security firm found over 35,000 business applications built with Grok AI publicly accessible without passwords, exposing sensitive data from 30 organizations including six Fortune 500 companies.

More than 35,000 business applications built with Grok AI were left publicly accessible without any login requirement. The exposure, discovered on November 10, revealed sensitive internal data from at least 30 organizations, including six Fortune 500 firms and confidential information from SpaceXSI and X.
Security researchers at Claroc uncovered the trove of exposed applications. Their investigation found patient medical data, employee passwords stored in plain text, and proprietary corporate records. One intensive care patient's lab results were among the compromised information. A separate hospital application contained a searchable roster of more than 3,500 staff members.
Applications built by SpaceXSI and X employees were also exposed. Twenty-four accessible apps contained confidential employee onboarding plans, sales territory maps, and advertising proposals. One sales map detailed 594 strategic accounts. Eight advertising proposals were exposed, including one marked confidential and intended for a U.S. federal agency. A timesheet application exposed 18 employee accounts with plaintext passwords, including administrator credentials.
Claroc estimates that between 650 and 800 of the apps held private data. Roughly one in ten applications contained internal business or personal information. About 2% of the apps held material rated as high or critical in severity. The firm is now notifying the affected organizations directly.
Root causes
The breach highlights critical default settings in AI app builder platforms. Grok's publishing feature automatically assigns each application a public grok.me web address. In a paid test account used by Claroc, the 'Anyone with the link' access setting was preselected by default. Researchers discovered the public links by scanning passive DNS records, which log app name lookups.
Accounts differ on the default visibility. Asaf Aprozper, Claroc's co-founder and CTO, stated that AI app builders often default to public visibility. A SpaceXSI representative countered that apps are private by default, set to 'Only you,' and require several deliberate steps, including acknowledging a legal notice, to be shared publicly.
These platforms enable employees without developer training to create software. A sales manager can build a customer dashboard. A healthcare administrator can generate a scheduling app. This bypasses traditional development, security reviews, and deployment approvals. The discovery shows a potential structural risk in the emerging market for AI-generated business software.
Adoption is accelerating. A September 2025 report found nearly 7% of U.S. enterprise CFOs had deployed agentic AI in live finance workflows. Another 5% were running pilots. This shift is driving increased security spending. Sixty-eight percent of financial institutions raised their cyber budgets year over year. IBM noted a higher percentage of companies planned to increase security spending after learning about frontier AI cyber capabilities.
Industry implications
The findings reveal systemic risks affecting businesses of all sizes. The problem is not unique to Grok. Applications built on platforms like Lovable, Bolt, and Replit have the same exposure risk because they also generate public URLs. In contrast, shared chats from AI assistants like Claude, ChatGPT, and Gemini keep secret link paths not visible to DNS scans.
This incident follows a pattern. Last month, the same research method exposed publicly accessible links for Meta's Muse. In July, Grok Build was found sending entire Git code repositories to xAI servers. The central question for business leaders is whether security governance can be embedded into the AI development process from the start, rather than applied after the fact.
AI makes software cheaper to build. The savings for a business, however, depend on how much survives the cost of securing what employees create. Claroc has asked SpaceXSI to make 'Only you' the default setting, move shared apps to hard-to-guess web paths, and warn builders about unsecured databases. After the report, SpaceXSI took 24 of the identified apps offline. The research did not establish that malicious actors exploited the exposed information. Claroc is notifying affected organizations directly.





