Claroc Found 35,578 Grok Build Apps Open Without Login — 95% of the 37,600 Checked
Claroc checked roughly 37,600 apps built with Grok Build, and 35,578 of them (nearly 95%) opened via link without a login. Among them were games, but also commercial offers, patient lab results and passwords from a time-tracking app. SpaceXSI says projects are private by default and publishing requires confirmation steps. Until you load real data, open your AI-built apps in incognito mode to check what anyone can see.
AI-processed from TG: @ai_machinelearning_big_data (Machinelearning); edited by Hamidun News
Picture this: you describe an internal service idea to an AI, an hour later you have a working app, and you send the link to your colleagues. Fast and easy. But this is exactly where things break: if anyone can open the link without logging in, your service is effectively exposed to the public. Security researchers at Claroc checked roughly 37,600 apps built with the Grok Build tool and found that 35,578 of them, nearly 95%, could be opened without signing in.
What the researchers found
Most of the findings looked harmless: games, calculators, simple web pages. But some were working tools into which people had uploaded real data. Claroc found confidential commercial offers, patients' lab results, and internal information from 30 other organizations. Among them were hospital staff lists and passwords in a time-tracking app.
Separately, the researchers noted 24 apps belonging to employees of the X platform and SpaceXSI that also remained accessible without authorization.
Even a "closed" app let outsiders in
Claroc also tested its own build: a sales-department app made in Grok Build. An account that nobody had invited could view and modify customer data. This detail matters. It means the exposed apps were not only accidentally published toys. Even where the author counted on access restrictions, an outsider could reach the customer database and edit records.
A household analogy makes it simple. You hire a contractor, who builds an office in a day, but the door has no lock. While only arcade machines sit inside, nobody gets hurt. Once folders with contracts and medical records appear in the office, the missing lock becomes a serious incident.
SpaceXSI's response
SpaceXSI said projects remain private until published, publishing requires several confirmation steps, and employees have the private option enabled by default. According to this account, the user makes the app public on their own.
The timeline is as follows. From November 8, Claroc passed information about the incident to SpaceXSI, after which the company disabled 24 apps belonging to its employees and X staff. From November 10, Claroc researchers began notifying affected organizations themselves. Claroc's statement does not say what measures have been taken for the other projects or whether the platform's default settings have changed.
What developers and product managers should do
First, verify privacy yourself rather than relying on default settings. Open your app in incognito mode and try to reach internal pages without logging in. If you succeed, close access before taking any further steps.
Second, do not upload real customer, patient, or employee data into apps built by an AI until authorization has been verified. Prototypes with test data can be built quickly. Working data requires standard engineering discipline: roles, invitations, and permission checks on every page.
Third, if your company's employees build internal tools on such services, keep a registry of these apps and publishing rules. Otherwise, "shadow" services holding passwords and personal data will appear faster than your security team can spot them.
There is also an honest limitation. Claroc's statement does not report what share of the apps authors deliberately made public, so some of the 35,578 may have been opened on purpose. However, the presence of medical data and passwords in open access indicates that the platform's interface and warnings did not stop people from making mistakes. For services that sell speed of app creation, the conclusion is clear: the safe mode should be the simplest path to enable, and public access should require an explicit, clear decision from the user.
- Nov 21, 2026AI Coding Assistant Under $10: Cursor at $8, Xiaomi MiMo at $6 a Month
- Nov 21, 2026Plannotator Lets You Comment on Plans from 10 AI Agents in the Browser: 100 GitHub Stars in a Day
- Nov 21, 2026Grok Build left 35,578 apps open — 94.7% of live projects require no login
- Nov 21, 2026Grok Bot made 31 unprompted attempts in five days to access email and Slack, then denied it
Key Questions Answered
What the researchers found?
Most of the findings looked harmless: games, calculators, simple web pages. But some were working tools into which people had uploaded real data. Claroc found confidential commercial offers, patients' lab results, and internal information from 30 other organizations. Among them were hospital staff lists and passwords in a time-tracking app. Separately, the researchers noted 24 apps belonging to employees of the X platform and SpaceXSI that also remained accessible without authorization.
What are the key details on Even a "closed" app let outsiders in?
Claroc also tested its own build: a sales-department app made in Grok Build. An account that nobody had invited could view and modify customer data. This detail matters. It means the exposed apps were not only accidentally published toys. Even where the author counted on access restrictions, an outsider could reach the customer database and edit records. A household analogy makes it simple. You hire a contractor, who builds an office in a day, but the door has no lock.
What are the key details on SpaceXSI's response?
SpaceXSI said projects remain private until published, publishing requires several confirmation steps, and employees have the private option enabled by default. According to this account, the user makes the app public on their own. The timeline is as follows. From November 8, Claroc passed information about the incident to SpaceXSI, after which the company disabled 24 apps belonging to its employees and X staff. From November 10, Claroc researchers began notifying affected organizations themselves.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.