xAI's Grok Build Uploaded Entire Git Repositories Without Permission

There's a special kind of dread that hits when you realize a tool you trusted has been quietly copying things it had no business touching. That's the situation a security researcher walked into with Grok Build, xAI's coding assistant. Instead of just reading the files it needed to help with a task, the tool was uploading entire repositories to xAI's servers, commit history and all.

Worse, it ignored the usual guardrails developers rely on. Files tagged with .gitignore, the standard way to tell a tool "hands off, this is private," got swept up anyway. So did .env files, which almost always contain API keys, database passwords, and other secrets that should never leave a developer's machine, let alone end up sitting on a third party's cloud in plain text.

This isn't a hypothetical privacy concern. Leaked API keys can mean drained cloud budgets, hijacked accounts, or worse, and commit history often contains old secrets that developers assumed were buried in the past. A tool designed to boost coding productivity instead created a fresh attack surface, and did it without telling anyone.

xAI's response was swift, at least on paper: the feature got disabled, and the company says every uploaded repo will be deleted. But swift after the fact doesn't undo the exposure window, and it doesn't tell developers how long this had been happening before someone caught it. For a company racing to compete with OpenAI and Anthropic on developer tools, shipping something that silently exfiltrates secrets is the kind of mistake that erodes trust fast, especially among the technical audience xAI needs most.