Highlights
  • Security firm Claroc published its findings on Nov. 10, 2026.
  • Of 43,551 apps loaded, 35,578 opened without sign-in.
  • That is 94.7% of apps that responded; only 1,993 asked for authentication.
  • Claroc estimates 650 to 800 apps contained business or personal data.
  • Cause: each app gets its own grok.me hostname, discoverable through passive DNS.

Building an app just by typing a prompt comes with a catch. Tens of thousands of apps made with Grok Build could be opened by anyone, no login required, and some held sensitive data.

Grok Build is SpaceXSI's AI app builder: describe an app and the AI writes and publishes it. The published apps turned out to be far more public than many creators realised.

This report draws on The Hacker News, Cybernews, The Decoder and Unite.AI, citing Claroc's research.

Quick answer: Claroc found 35,578 Grok Build apps open without login, 94.7% of those that responded. Each app is published to its own grok.me hostname meant to be unguessable, but the hostnames show up in passive DNS data. An estimated 650-800 apps held business or personal data, including medical data and plain-text passwords.

What Claroc Found

Claroc loaded 43,551 distinct Grok Build apps. 35,578 opened without sign-in, while 1,993 returned an authentication error.

It estimates 650 to 800 apps contained private business or personal data. Outlets differ on the share, from about 2% with high sensitivity to roughly one in ten. Reported exposures include SpaceXSI's own sales plans, employee rosters, medical data and plain-text passwords.

Read Also: Hackers Used the Hermes AI Agent to Steal 2,000 Tajik Government Files

Why the Apps Were Exposed

Grok Build publishes each app to its own hostname under grok.me. The links were meant to be unguessable, but they appear in passive DNS records, archives of domain lookups that anyone can query.

The Decoder reported that the publish dialog on Claroc's test account preselected "Anyone with the link", so many creators likely never knew their apps were public.

Key Numbers

From Claroc's research:

MetricFigure
Apps loaded43,551
Open without login35,578
Asked for authentication1,993
Share open (of responding)94.7%
Estimated with confidential data650-800
Staff apps taken down by SpaceXSI24

Source: Claroc findings as reported by The Hacker News and Cybernews, November 2026.

SpaceXSI's Response and Lessons

Claroc says it reported the issue on Nov. 8 and 9. SpaceXSI then removed 24 staff-built apps but not other users' apps, and had not replied as of Nov. 10.

For anyone using AI app builders:

  1. Check sharing settings every time you publish; never trust the default.
  2. Do not treat a hard-to-guess link as security. Add a login for apps with data.
  3. Keep real customer data, medical records and passwords out of test apps.
  4. Take down apps you no longer use.
  5. Companies should set rules on which AI tools staff may use.

FAQ (Frequently Asked Questions)

1. How many Grok Build apps were exposed?

Claroc found 35,578 apps open without login, 94.7% of those that responded.

2. Why were they exposed?

Each app's grok.me hostname could be discovered through passive DNS data and opened without signing in.

3. What data leaked?

Reports cite SpaceXSI sales plans, employee rosters, medical data and plain-text passwords.

4. When was it reported?

Claroc notified SpaceXSI on Nov. 8-9 and published on Nov. 10, 2026.

5. How can I keep AI-built apps safe?

Check sharing settings, add a login, avoid real data in tests and remove unused apps.

Conclusion

Claroc's findings show the downside of instant AI app building: tens of thousands of apps were public because their only protection was an obscure address.

AI can build an app in minutes, but protecting the data in it remains the creator's job.