SpaceXSI's Grok Build coding agent exposes sensitive data from 35,000 apps

Covered by 2 sources
Read onArs Technica· 2 min
SpaceXSI's Grok Build AI coding agent leaked sensitive data from over 35,000 apps due to a flaw in link handling that made them publicly discoverable. The exposure included internal documents from SpaceXSI and other organizations, contradicting claims that apps remain private unless shared intentionally. Following disclosure, SpaceXSI and affected companies acted to address the vulnerability.

Also covered by 1 source

What happened

A security researcher discovered that Grok Build, an AI coding service owned by SpaceXSI, made over 35,000 apps publicly accessible due to a flaw in its link generation system. The issue allowed sensitive data from SpaceXSI and other organizations to be exposed, including internal documents and confidential information. The links to these apps could be found through certain DNS services, making them accessible to anyone who searched. After the exposure was disclosed, SpaceXSI and some affected organizations took steps to address the vulnerability.

Key details

  • Over 35,000 apps created with Grok Build were exposed by default.
  • Sensitive data from SpaceXSI, a hospital group, a fire department, and several Fortune 500 companies was leaked.
  • The flaw involved Grok Build placing random link phrases to the left of the domain, making them discoverable by passive DNS servers.
  • Only one exposed app required a password for access.
  • SpaceXSI removed 24 of its own exposed apps after being notified publicly.

Why it matters

Organizations using Grok Build, including large companies and public sector groups, faced the risk of confidential information being accessed by unauthorized parties. This exposure could lead to data breaches, reputational harm, and regulatory consequences for those affected. Companies now need to review their use of Grok Build and address any remaining vulnerabilities to prevent further leaks.


More in Cybersecurity