Digital Event Horizon
Grok Build coding agent leaks confidential data, including proprietary business information and personal identifiable information, due to a design flaw that allows users to share apps with anyone who knows the exact URL. The incident raises concerns over the security of capability-based access control and the need for more stringent measures to prevent data exposure.
Grok Build, a coding agent, has leaked sensitive information from thousands of its users.The leak was caused by a design flaw in the agent's capability-based access control.The randomness mechanism used by Grok Build was easily enumerable using passive DNS services.Over 35,000 links ending in grok.me were public, exposing sensitive data.Users had to change the sharing setting to prevent data exposure, which was not always enforced.The issue highlights the need for more stringent measures to prevent data exposure in AI platforms.
Grok Build, a coding agent developed by SpaceX, has been found to be leaking sensitive information from thousands of its users, including proprietary business data and personal identifiable information. The leak was discovered by security firm Claroc, which revealed that the agent's design flaw allows users to share their apps with anyone who knows the exact URL, effectively exposing sensitive data to the public.
The issue lies in the way Grok Build implements capability-based access control, a technique designed to make URLs easy to share but hard to guess. However, the randomness mechanism used by Grok Build is positioned to the left of the domain, making it easily enumerable using passive DNS services. This allows attackers to capture the transaction and access the sensitive data stored in the app.
According to Claroc, more than 35,000 of the 43,601 links ending in grok.me were public, meaning that clicking on the link opened the corresponding app and all data it stored. This is despite the fact that the representative of Cursor, the AI coding service SpaceXSI recently acquired, stated that Grok Build apps are private by default and nothing goes public unless the creator chooses to share it.
In fact, the representative's statement was contradicted by Asaf Aprozper, the co-founder and CTO of Claroc, who stated that users have to change the setting from "Anyone with the link" to "invited users" or "only you" to prevent the data from being exposed. Aprozper's tests showed that this setting is not always enforced, leaving users vulnerable to data exposure through shared links.
The issue is not limited to Grok Build, as other AI platforms, including Lovable, Bolt, and Replit, also give users the ability to generate shareable links. However, Claroc found that these platforms follow the correct sharing link practice, with secret randomness in the path, making it difficult for attackers to access the sensitive data.
The incident raises concerns over the security of capability-based access control and the need for more stringent measures to prevent data exposure. SpaceXSI has removed the links to the affected apps, but the incident highlights the importance of carefully designing and implementing access control mechanisms to prevent similar incidents in the future.
Related Information:
https://www.digitaleventhorizon.com/articles/Grok-Build-Coding-Agent-Leaks-Confidential-Data-Raises-Concerns-Over-Capability-Based-Access-Control-deh.shtml
https://arstechnica.com/security/2026/11/grok-build-coding-agent-leaks-oodles-of-confidential-info-some-belonging-to-spacexsi/
Published: Tue Nov 10 17:25:06 2026 by llama3.2 3B Q4_K_M
