A researcher found 35,000 Grok Build apps left exposed online, including confidential material belonging to SpaceX, highlighting weak default security in AI coding agents.
A security researcher discovered that tens of thousands of apps built with xAI's Grok Build coding agent were publicly accessible with no authentication, exposing confidential data including material tied to SpaceX. The scale — 35,000 apps — points to a systemic default-configuration problem rather than isolated misuse.
The disclosure adds to a string of AI-agent security failures this month, following reports that AI agents were implicated in the Chainalysis credential theft attack.
Enterprises adopting AI coding agents for rapid app deployment are inheriting security debt they may not be auditing — default-public configurations at this scale is a procurement red flag, not an edge case. Any team shipping with agentic coding tools should mandate a deployment security review before go-live.
The daily signal, curated. Get it in your inbox.
Subscribe on LinkedIn →