Grok Build Security Flaw Exposes Sensitive Data From Thousands of Users and SpaceXSI

“We were shocked,” security firm Claroc told Ars Technica, describing how thousands of applications built by the Grok Build AI tool exposed confidential data online.

The Tech TL;DR:

• Security researchers found over 35,000 Grok-built applications exposed online without required authentication checks.
• Leaked files included internal business documents from SpaceXSI, hospitals, fire departments, and Fortune 500 firms.
• Passive DNS services indexed the subdomains, allowing unauthorized users to enumerate live projects.

When creators publish an application, the platform assigns it a unique subdomain under grok.me. Although developers intended for random four-word descriptors to secure these URLs, placing that random string to the left of the domain allows passive DNS lookup services to capture and index the addresses.

Claroc discovered 37,571 Grok-built apps online during their research, finding that 35,578 of them permitted access to anyone without requiring a user sign-in. Approximately 10 percent of those open instances contained sensitive personal or internal corporate material. Among the exposed data, researchers identified medical exam results, galley checklists from a cruise line, intensive care patient briefings, and an active fire department shift board.

Irony marked the discovery when investigators found 24 internal applications created by staff at SpaceXSI, the company that develops the Grok platform itself. Biz & IT – Ars Technica reported that these exposed corporate assets included a confidential onboarding plan for a new enterprise sales employee, a sales territory map detailing 594 accounts, internal system addresses, and advertising proposals.

“The company that makes Grok had its own internal material open to anyone holding the link,” Claroc wrote in a blog post.

Following notification via its security.txt file on Sunday, SpaceXSI removed all 24 of its internal staff applications. However, the company left the tens of thousands of other vulnerable user applications active on the platform. Researchers estimated that roughly 2% of all open apps contained material serious enough to rate as high or critical severity.

To verify access controls, security analysts deployed two test applications using default settings. Testing revealed logic flaws where external accounts bypassed membership validation checks and modified database records without authorization.

Always consult with certified IT and cybersecurity professionals before altering enterprise networks or handling sensitive data.