Grok Build’s AI-Generated Apps Leaked SpaceX’s Internal Files to Anyone With a Link

Security researchers found over 43,000 Grok Build apps using passive DNS, and nearly 95 percent were open to anyone with no login, including 24 of the apps built by SpaceXSI itself.

Get a summary in:
Hacker binary attack code on screen, representing the Grok Build data leak
Follow Teck Hustlers on Google
Add Teck Hustlers as a preferred source to see more of our stories in Google Top Stories.
Add as preferred source on Google

TL;DR: Security researchers used passive DNS data to find more than 43,000 live apps built with Elon Musk’s Grok Build coding tool, and nearly 95% of them opened to anyone with the link, no login required. About one in ten held sensitive business data, and the worst exposure belonged to SpaceXSI, the company behind Grok Build itself, which left sales pipelines, advertiser budgets and employee lists sitting in public apps for days.

Server racks with network cables, representing exposed cloud apps

A vibe-coding shortcut turned into a search engine for secrets

Grok Build lets anyone describe a web app in a chat window and get a working, published version back in minutes. That’s the whole pitch: no setup, no deployment pipeline, just a prompt and a live URL at something like fourwords.grok.me. It’s the same vibe-coding pattern Google, Replit and a handful of other players have been racing to ship, and on its own that’s not a problem. The problem is what happens after the app goes live.

Every Grok Build app gets published to a URL built from a handful of randomly chosen words. Security researchers at Claroc found that because Grok Build puts those random words in front of the domain name, rather than behind it the way Google’s Gemini does it, passive DNS logs end up recording the full address every time someone’s device queries it. Passive DNS services exist specifically to catch and index domain names seen in real traffic, and once an address lands in one of those databases, it’s searchable by anyone. Claroc pulled 43,601 grok.me addresses this way. No breach, no exploit, just a list that was never supposed to be public sitting in a public index.

From there the second problem took over. Grok Build’s default publish setting is “Anyone with the link,” ahead of the more restrictive “Invited Users” and “Only You” options. Claroc found that 35,578 of those 43,601 apps, nearly 95%, loaded with zero authentication. Only one app in their entire sample asked for a password, and that one turned out to be an archive of ad-measurement studies for a streaming service.

What was actually sitting out in the open

Roughly one in ten of the apps Claroc checked contained sensitive or internal business information, and the firm rated about 2% of those as high or critical risk. The examples they published aren’t abstract. A hospital had an employee directory with more than 3,500 names exposed. An intensive care unit’s patient briefing was readable by anyone. A trucking company’s recruitment database listed over 1,000 people. One employee timesheet app shipped with downloadable source code that included 18 accounts with plaintext passwords, administrator credentials among them. A fire department’s app was on the list too. Claroc also said six of the exposed organizations are Fortune 500 companies and 26 have more than 1,000 employees.

To show how little it would take for someone to actually abuse this, not just view it, Claroc built two test apps of their own. One let an unauthenticated visitor edit a shared contact database outright. The other accepted a sign-in from any Google account, not just people on the team that built it. Neither required any real effort to set up that way; it’s just where Grok Build’s defaults land if nobody changes them.

The irony: SpaceXSI leaked its own business the same way

Grok Build is built by SpaceXSI, the company formed when SpaceX and xAI merged their operations. Claroc found 24 of SpaceXSI’s own apps sitting wide open using the exact same flaw they’d just documented in everyone else’s. What was in them reads like an internal sales binder: an onboarding plan, employee name lists, a sales territory map covering close to 600 strategic accounts with live pipeline details, advertiser decks broken out by budget tier, and seven campaign performance reports for advertisers around the world. One of those budget tiers, tied to a US federal agency, was marked confidential. Separate reporting from Engadget lists eight X advertising proposals, four live advertiser account boards and two copies of an X brand-safety playbook among the exposed material.

Claroc’s reaction, in their own words: “We were shocked” that “the company that makes Grok had its own internal material open to anyone holding the link.” It’s a fair gut punch. A company building a product that generates publicly reachable apps by default didn’t catch that its own team was using it the same risky way as everyone else.

SpaceXSI fixed its own exposure. The rest of the platform is still wide open

Claroc says it notified SpaceXSI on November 8. TechRadar reports that within two days, SpaceXSI’s apps were no longer reachable. Engadget’s account of how that happened is less flattering: it reports SpaceXSI didn’t respond until Claroc publicly tagged Elon Musk and another company official on X, and that the fix only covered SpaceXSI’s own 24 apps. Claroc wrote that SpaceXSI “has not done the same for the apps other Grok users and organizations left open.” Some of the other companies named in the research moved fast on their own once they learned they were exposed. Most of the tens of thousands of other apps Claroc catalogued are presumably still sitting there, discoverable the same way, with the same default settings. Neither xAI nor SpaceXSI has issued a public statement on either the naming scheme that makes apps discoverable through passive DNS or the default sharing setting that makes them openly readable once found.

This is the same shape of problem that’s followed AI coding and agent tools all year: a feature built for speed ships with the permissive option as the default, and the people using it never think to change it because nothing in the product nudges them to. We covered that same dynamic last week when AI agents were found handing over secret keys to other agents that simply asked nicely, and again with the Railway breach where attackers chained through stolen credentials to rack up $740,000 in fraudulent AI usage. Different platforms, same root cause: convenience defaults that nobody revisits until a researcher goes looking.

The bottom line

Claroc didn’t need to hack anything to pull this off. They read public DNS records and clicked links that were never locked in the first place. That’s the part that should worry anyone using a vibe-coding tool to spin up something quick: the barrier between “working demo” and “world-readable database” is one unchanged dropdown menu. SpaceXSI closing its own 24 apps days after getting publicly called out isn’t a fix, it’s damage control. Until Grok Build changes its default sharing setting and stops leaking addresses through DNS, the tens of thousands of other apps on the list are exactly as exposed as SpaceXSI’s were, and nobody building with it right now has any way of knowing if they’re one of them.

Follow Teck Hustlers on Google
Add Teck Hustlers as a preferred source to see more of our stories in Google Top Stories.
Add as preferred source on Google
About

The Teck Hustlers Editorial Team covers technology, business, finance and gaming news for readers who want the story behind the headline. Every article is researched, written and edited by our in-house team, checked against company announcements, filings and reputable reporting, and updated when new facts emerge. We focus on what each development means for founders, professionals, investors and everyday users. Have a tip or a correction? Reach us through our contact page.

Up next: Meta Puts Its AI Companion Muse on Your Wrist, Two Months After Launch