Nearly 35,600 apps built with Grok’s app builder have been found publicly accessible—revealing confidential business records, personnel details, patient data, and even plaintext passwords without any hacking involved. The issue stems from app instances created using Grok’s Build mode under the grok.me domain, where the published apps defaulted to unrestricted public access.
How the Exposure Was Discovered
On November 8, 2026, researchers from cybersecurity firm Claroc analyzed more than 43,500 app addresses under grok.me via passive DNS—a service that logs domain names as they’re looked up. Of those, 37,571 apps were still active. Shockingly, 35,578 of them (about 94.7%) allowed full access to anyone, while only 1,993 required sign-in. These aren’t guessable URLs: each app’s domain is registered automatically and becomes publicly searchable once someone visits the address. That means all 35,578 fully open apps were discoverable online without trickery.
What Was Exposed
Most of the open apps were fairly innocuous—games, calculators, landing pages. But roughly one in ten held sensitive internal or personal business data, and nearly 2% implicated serious high or critical-severity risks. Among the revealed apps were:
- An intensive care unit dashboard showing patients’ lab results and ventilator status;
- A hospital system’s roster with over 3,500 staff members;
- A fire department shift board listing badge numbers;
- A contractor’s timesheet app that stored 18 staff accounts with passwords written in plain text and included in code that any visitor’s browser downloaded.
The data exposure also extended to apps belonging to SpaceXSI (formerly xAI), the parent company of Grok. Its own team had 24 apps set to open access—items such as an onboarding plan, advertising proposals, and a sales map intended for use by a U.S. federal agency. In total, Claroc identified 30 recognizable organizations with exposed internal apps; 26 of them employing more than 1,000 people and six being Fortune 500 firms.
Developer Settings and Company Response
Claroc reported that even paid Grok accounts defaulted to public link access, meaning apps remained publicly viewable and editable without login protection. One app using Google Sign-In checked that users were logged in but did not verify whether they were part of the intended team—allowing any valid Google account to read, modify, or delete customer records. When questioned, Grok’s parent company claimed that apps are meant to stay private until the creator explicitly publishes them. Claroc countered that all the exposed apps did, in fact, go through the official publishing workflow.
Following the disclosure, SpaceXSI took down the 24 exposed apps that it owned, but the vast majority—those belonging to other users—remain live and vulnerable.
Broader Context & Advice
This isn’t Grok’s first privacy misstep. In 2025, more than 370,000 shared Grok chat sessions were indexed in Google search due to misconfigured privacy settings. Claroc’s full research offers more detail on how DNS records and app naming expose content inadvertently.
To reduce risk, organizations are advised to audit any Grok or similar AI app builder tools being used by staff. Check whether apps are publicly accessible, test login restrictions with outside accounts, and review app code—especially script files—for passwords or internal endpoints. Any grok.me app that isn’t actively managed or owned should be unpublished; doing so takes the app offline within seconds.
What this means: AI app builders promise speed and ease but may come with significant security trade-offs. This incident underscores how default configurations and oversight gaps can create patterns of mass data exposure. Companies and developers using AI tools need to be far more vigilant about access controls, publishing workflows, and settings—because data leaks like this become much harder to contain once indexed or otherwise public.
