How Grok Build's Default Sharing Settings Exposed 35,000 Apps

When security firm Claroc investigated the Grok Build coding agent, they uncovered a massive privacy exposure: over 35,000 applications generated by users were left wide open to anyone capable of enumerating live project links. According to Ars Technica, the leaked data included internal business strategies, medical exams, employee identifiers, and proprietary materials belonging to major organizations, including SpaceXSI itself.
Where did the architecture fail?
The core vulnerability stems from how sharing links are constructed. While platforms like Google, Claude, and ChatGPT place the randomized security tokens within the URL path—making them invisible to passive DNS lookup servers—Grok Build placed the generated words to the left of the domain. This architectural choice allowed passive DNS services to easily map and index the active project domains.
Furthermore, Claroc's findings directly contradict statements from a Cursor representative, who claimed apps are private by default. Tests showed that Grok Build defaulted to public access under the "Anyone with the link" setting. Out of more than 43,000 discovered links, nearly 95 percent exposed their underlying databases without requiring authentication.
How to protect your work
If you use AI coding tools that generate shareable web apps, you must take manual precautions before deploying anything containing sensitive information.
- Check default permissions: Always verify that link-sharing settings are toggled away from public access.
- Restrict invites: Manually change project sharing permissions to "invited users" or "only you" before inputting any proprietary data.
- Audit active links: Regularly review external deployments and avoid pasting internal business logic into rapid prototyping agents without authorization.
Verdict: Who should (and who shouldn't) use it
Grok Build remains a powerful tool for rapid prototyping, letting users spin up functional software in minutes through text or speech. However, this incident proves that treating AI-generated apps as secure by default is a dangerous gamble. If your workflow involves handling proprietary code, customer lists, or internal metrics, verify your platform's privacy controls immediately or avoid using public link-sharing features altogether.
Sources
Frequently asked questions
- What caused the Grok Build data leak?
- A combination of default public sharing settings and a URL structure that placed randomized tokens to the left of the domain, making them discoverable via passive DNS.
- What kind of data was exposed?
- The exposed apps contained internal business plans, employee identification numbers, healthcare rosters, sales pipelines, and aerospace logistics data.
- How can users prevent similar leaks?
- Always manually change project permissions from public access to restricted invite-only modes before creating or sharing apps with AI coding agents.
AI Briefing in 3 Minutes
Get what actually matters in AI & tech delivered free to your inbox. Zero spam, 1-click unsubscribe.
Comments
0 commentsDeixe seu comentário
Be the first to comment.
Continue Reading

Gemini Robotics 2.5: Why Industry Still Can't Trust AI Yet
Gemini Robotics 2.5 hits 96% factory success

Apple Starts Charging $50 for Extra AI Usage: What You Need to Know
Apple launches 50 dollar AI usage packs

Are Claude Code Settings Secretly Ditching Your Passwords on GitHub?
Claude Code leaks secrets to public GitHub