Signal

Researchers find that Grok Build is leaking data of over 35,000 apps via shareable links, including proprietary information, some belonging to SpaceXSI

First reported by Ars Technica ·

The signal ●●●○ Compiled by AI from Ars Technica, Techmeme, Claroc Blog and RuntimeWire
Why you might care

The data security of your AI-generated applications can no longer be assumed to be private by default.

What happened

A security research firm, Claroc, has discovered that Grok Build, an AI coding agent from SpaceX, is leaking sensitive data from over 35,000 applications through its shareable link feature. The exposed information includes proprietary business plans, advertising proposals, medical records, and employee identification numbers. Notably, some of the leaked data originated from 24 applications developed by SpaceX itself. The vulnerability stems from the design of Grok Build's shareable links, which, unlike similar features on platforms like Google Docs, place the random URL components to the left of the domain name. This allows passive DNS services to easily enumerate and access these links, effectively making them public. Claroc identified the issue and notified SpaceX, which subsequently began removing the exposed links for its own applications. Other companies, including a Peruvian construction firm, Tesla, and a US Navy contractor, also had sensitive data exposed.

What it means

The core issue lies in Grok Build's implementation of capability-based access control, where link possession is intended to grant authorization. However, by placing the random URL segments before the domain name, Grok Build inadvertently makes these links discoverable through passive DNS logging, a method typically used to track DNS lookups. This design flaw negates the intended privacy of randomly generated URLs, exposing a vast number of applications and their associated data to anyone who knows how to search these public DNS records. The broad scope of exposed data, ranging from internal sales strategies to employee PII, highlights a significant security oversight with potentially far-reaching consequences for businesses and individuals.

This incident underscores a critical security gap in the rapid development and deployment of AI-powered tools, particularly those offering simplified sharing mechanisms. While Grok Build's creators and some competitors offer options to restrict link access, the default setting and the underlying vulnerability mean that many users may be unknowingly exposing confidential information. The situation necessitates a re-evaluation of security best practices for AI development platforms, emphasizing robust default privacy settings and secure URL generation techniques that prevent enumeration. Users of such platforms must now be hyper-vigilant about checking and configuring sharing permissions to avoid similar data leaks.

AI-written summary. May contain errors.

Space