AI StrategyChoosely EditorialEarly assessment

Grok Build exposure: who can open your app?

A report of accessible Grok-built apps raises a practical question: can your intended users get in while everyone else stays out?

← Back to Latest
Editorial illustration of a personified Grok symbol holding a link-shaped pass beside an open access gate, while Choosely Chimp checks a list of users.

If you have used Grok to build an internal dashboard or client portal, review who can open the published app before adding sensitive work. Then have its responsible developer check who can access the underlying records. A working login screen is only part of that answer.

What Claroc found

In its November 10 investigation, security company Claroc reports that 35,578 of the 37,571 still-online Grok apps in its discovered-address sample opened without sign-in. Most were games, landing pages and calculators. Some held internal business or personal information, it says.

Claroc also reports 24 open apps built by staff at X and at SpaceXSI, Grok’s maker, formerly called xAI.

This is an early assessment based on public reporting: Choosely has not inspected affected apps or reproduced the tests, and Claroc’s discovered-address sample is not representative of all Grok apps. Claroc also sells services for securing AI-built apps.

A link can travel further than you expect

The official Grok Build announcement describes published app addresses and owner-selected audiences, including anyone with the link.

An obscure address can still become findable. Claroc says it found the addresses using common subdomain-discovery techniques, and nobody had sent it any of them. It points to passive DNS, which records hostname lookups, as one way such names become searchable. DNS is the system used to look up a hostname so a browser can reach its server. The IETF’s DNS privacy guidance explains how those observations can be collected and shared.

For an owner, the implication is simple: choose access on the assumption that someone outside the intended audience could learn the address. Keeping a link out of a public post does not answer who can use it.

Why the login screen mattered in Claroc’s test

Claroc built two apps with made-up data. In its sales-team example, Grok added Google sign-in, but the generated code enrolled whoever signed in before checking team membership. An uninvited Google account could see all the test customer records, edit one and delete another, Claroc reports.

That is a concrete example of the distinction OWASP makes between authentication and authorization. Authentication establishes who someone is. Authorization determines what that person may see or change. In Claroc’s test app, sign-in identified the visitor, and the code then let them in anyway.

The practical check belongs with the responsible developer: verify permission at both the page and the data layer, including which records each user may access. OWASP recommends server-side checks on every request. The observed result belongs to Claroc’s own test app; the prevalence of that behavior elsewhere is unknown.

How the company responded

Ars Technica attributes an emailed response to a Cursor representative. Cursor was acquired by SpaceX in August. The representative said Grok Build apps are “private by default” and that nothing goes public unless the creator chooses to share it.

Ars says the representative also pointed to competing builders’ shareable links and did not answer follow-ups about intent, removal or customer notifications.

Separately, Claroc recounts a response SpaceXSI gave to a reporter: publication takes several steps, employee accounts default to Only you, and the process involves more steps than Lovable’s.

Claroc agrees that nothing becomes public before publication. Its dispute concerns the next step: in the paid account it tested, the audience choices included Anyone with the link, Invited users and Only you, with Anyone with the link preselected.

The official announcement’s illustrated publish sheet also displays Anyone with the link alongside Not published. Current defaults across account types, completed remediation and notification coverage remain unverified.

What owners should check now

Start with apps and data you own or are authorized to manage. You need an agreed audience, an accountable owner and permissions that match the work.

Your situationNext stepWho owns it
You are preparing an app for sensitive work.Identify its intended users and data. Use nonsensitive example records until access is verified.App owner and data owner
Sensitive work is already in an app shared with anyone holding the link.Narrow the audience to the smallest setting that still works for intended users, or unpublish if the app can go offline. If others rely on the app, agree the change with whoever runs it first, then start the incident steps below.App owner; administrator or security lead where applicable
The app displays a login screen.Verify that pages and underlying data reject unauthorized users and enforce each user’s permitted records and actions.Responsible developer
Nobody can explain who owns the app or its permissions.Assign an owner and resolve the access requirement before adding more sensitive work.Team manager and responsible developer

Claroc’s own controls test provides a useful detail: setting its first app to Only you returned HTTP 401, an access rejection, for both the page and its data function when requested by a stranger. Unpublishing its second test app took it offline within seconds. For an existing business app, the administrator and security lead should agree on containment that protects both the data and the service.

In its November 10 account, Claroc said SpaceXSI had taken down the 24 staff-built apps, while other users’ apps remained online. It had emailed three affected organizations at publication and said further notices were going out. Owners should review their own access without waiting for a notification.

If exposure is suspected, involve the security or privacy lead and preserve necessary evidence through the organization’s incident process. NIST’s guidance calls for coordinated response and protected incident records.

Where credentials were exposed, the responsible team should revoke or rotate them and coordinate cleanup. GitHub’s sensitive-data guidance illustrates the limits of removing already distributed material. Applied here, taking an app offline cannot establish that no earlier copies exist.

Choosely’s verdict

Before putting sensitive work into a Grok-built app, establish who should get in and have the responsible developer verify what those people can access. If either answer is unclear, keep sensitive data out while the owner resolves it.

Claroc’s test shows why that second question matters: a login can identify someone and still admit them to a team they were never meant to join. Your app’s permissions need to work even when someone else finds its link.

The Change Brief

Get the week’s AI changes in one clear read

Pricing moves, tool launches, free-tier changes and practical stack updates, filtered for people who actually use these tools.

Stay ahead of AI without following it all day. We’ll send you what matters each week.

Continue reading

Related reads