Tech News · Grok
Grok Build apps exposed — what businesses should do now
“AI app builders turned publishing into a few clicks, and the default answer to who can see it is everyone.”
— Asaf Aprozper, Claroc Research
What happened
Grok Build apps were exposed at scale: security firm Claroc reported on November 10, 2026 that 35,578 apps built with Grok, the AI assistant from SpaceXSI (the company formerly called xAI), opened for anyone on the internet without a sign-in. Claroc checked 43,551 app addresses under the grok.me domain, found 37,571 still online, and 94.7% of those loaded for a stranger; only 1,993 returned the “not authorized” error a private app should give.
Most of the open apps were harmless games, calculators and landing pages. But Claroc estimates about 1 in 10 held internal business or personal data, and roughly 650 to 800 held private data. Examples the researchers confirmed by hand include a hospital group’s searchable roster of more than 3,500 employees, an intensive care briefing on a single patient, a fire department’s shift board, and a contractor’s timesheet whose code contained 18 staff accounts with plain-text passwords, admins included. Twenty-four of the open apps were built by SpaceXSI and X employees themselves, among them a sales territory map of 594 accounts and an advertising proposal marked confidential. Beyond those, data from 30 other identifiable organizations was exposed, 26 of them with more than 1,000 employees and six in the Fortune 500.
Why it happened matters for every business using AI app builders. Each published Grok app gets its own web address (a subdomain of grok.me). When anyone opens that address, the lookup is recorded by passive DNS services that can be searched, so “secret” links became findable lists. In the paid account Claroc tested, the publish screen had “Anyone with the link” already selected. In two test apps the researchers built with fake data, Grok added a database with no login to one, and in the other wrote a team sign-in that let any outside Google account in to read, edit and delete customer records.
According to the researchers, SpaceXSI said Grok apps stay private until their creator chooses to share them and that employee accounts default to “Only you.” The company took down its 24 staff apps after the report, but had not done the same for other users’ open apps. The Hacker News notes the chat-based builder was opened to every Grok plan on August 19.
FuzeForth’s take
This is not really a Grok story. It is a story about how business software now gets made. A sales manager, an office admin or a shift supervisor can describe a tool in a chat window and publish it in minutes, with no IT ticket and no review. That speed is real value, but it also means customer lists, staff rosters and pricing sheets can end up on a public link nobody in leadership knows exists. Grok, Lovable, Replit, v0, Bolt, Base44 and similar AI app builders all make publishing easy; the difference is in their defaults.
Small and mid-size businesses are the most exposed, because they rarely have a security team watching for this. If your team uses any AI app builder, assume at least one internal tool is shared more widely than intended until you have checked. Treat an AI-built app that holds customer, patient, employee or financial data the same way you would treat a new piece of business software: someone owns it, it requires a real login tied to your company, and it gets reviewed before it goes live.
For anything that runs core operations, such as CRM, scheduling, invoicing or an ERP workflow, a properly built business software platform with roles and company sign-in is the safer foundation, and AI can still sit on top of it. If you want AI doing real work, look at AI agents built with access controls rather than one-off public apps. And if you are unsure what is already out there, an ongoing website and security care plan can include a regular audit of public-facing tools.
Key points
- Claroc found 35,578 of 37,571 live Grok-built apps (94.7%) open to anyone without a sign-in.
- About 1 in 10 open apps held internal or personal data, including employee rosters, patient details and plain-text passwords.
- App addresses were discoverable through passive DNS records, so a hard-to-guess name was not real protection.
- In the account tested, “Anyone with the link” was preselected at publish; ChatGPT’s builder defaulted to “Only me” in the same test.
- SpaceXSI removed its own 24 staff apps but, per the researchers, not the apps of other users and organizations.
Frequently asked questions
Are Grok Build apps private by default?
SpaceXSI says apps stay private until the creator publishes and shares them. In Claroc’s paid test account, though, the publish screen proposed “Anyone with the link” as the access setting, and employee accounts were the ones set to “Only you.”
How do I check if my company’s Grok apps were exposed?
Ask staff which AI app builders they use, then open each published app in a private browser window while signed out. Any app with internal data should be set to “Invited users” or “Only you” and should show an error page to a stranger; unpublish anything nobody owns.
Does this affect other AI app builders like ChatGPT or Lovable?
Any builder that gives each app its own web address can have that address show up in DNS data. What protects you is the access setting, so check the sharing defaults and login rules in every tool your team uses, not just Grok.
