Security firm Claroc finds Grok Build shared-app links enumerable via passive DNS, exposing data from thousands of apps including 24 SpaceXSI internal apps
Edited by Elliot Marsh, Editor · written with AI from the linked sources (how)
Claroc published research showing Grok Build app links (random words placed to the left of grok.me) can be enumerated through passive DNS. It found 43,601 links, over 35,000 of them public, exposing business, medical and employee data from many organizations. 24 of the apps were SpaceXSI's own. After Ars Technica emailed questions, SpaceXSI removed the links for those 24 apps, which now return 'app not found'. Most other exposed apps remained available at publication.
Entities: Claroc, SpaceXSI, Grok Build, Elon Musk, Matt Palmer, Asaf Aprozper
0 primary
What happened
Security firm Claroc reported that Grok Build shared-app links, which put random words to the left of grok.me, can be found through passive DNS records. It says it found 43,601 links, over 35,000 of them public, exposing business, medical and employee data. 24 of the apps belonged to SpaceXSI itself. After Ars Technica sent questions, SpaceXSI removed those 24 links (they now return "app not found"), but most other exposed apps were still reachable at publication.
Why it matters
Anyone who built and shared an app on Grok Build may have had its contents discoverable by third parties, so affected teams should assume exposure and review what those apps held. The flaw shows that secret links only work if the secret never leaves the page, and putting it in a hostname leaks it to DNS logs. Buyers of AI app builders have a concrete question to put to vendors. The reported mention of Fortune 500 and defence contractor data is serious, but the article names no organisations and the claim cannot be checked from what is given.
What is noise
The figures and the Fortune 500 and defence contractor claims come from Claroc's own research as relayed by Ars Technica, with no independent verification or sample of the data shown. The comparison with Lovable, Bolt and Replit is a risk suggestion, not a demonstrated leak. Also, "public" links being exposed is partly the point of public links; the real harm is data that owners believed was private or obscure. The headline's "oodles" adds colour, not information.
Watch next
- 01Whether SpaceXSI changes the link design (for example moving the random token into the URL path) or removes or restricts the remaining exposed apps, and whether it issues a public statement or advisory.
- 02Whether Claroc publishes its methodology or technical write-up, or any independent researcher reproduces the 43,601 count, and whether named organisations confirm their data was exposed.
- 03Whether Lovable, Bolt, Replit or others are shown to have the same hostname-based exposure, or publicly confirm they do not; also any regulator or data-protection body response over the medical and employee data.
Coverage
2 storiesNewsletter
Get the no-hype AI news straight to your inbox
Weekly. No spam, unsubscribe any time. Privacy
More infrastructure signals
Full feed →- Deepseek releases V4.1-Flash, an open-source model that sharply cuts KV cache memory and input-processing compute for AI agents10 Sept 202682
- Anthropic launches Nasdaq IPO roadshow, pricing 825 million shares at $70-$78 for up to $90.4 billion total raise at up to $1.02 trillion valuation10 Nov 202680
- Mistral releases Mistral Large 4, a ~1T-parameter open-weight multimodal MoE model, with day-0 Hugging Face support29 Oct 202680
- Anthropic files public S-1 with SEC for Nasdaq IPO under ticker "ANT"26 Oct 202680