Grok Build is hit by a hack that steals users’ data: the attacker uploaded the “entire home directory” to the cloud, shocking developers—everything has been leaked

Have AI code-writing tools turned into privacy vacuum cleaners? Multiple developers on the X platform have warned that Grok Build, xAI’s development tool under Elon Musk, has raised serious concerns about privacy leakage. A user posted an accusation, saying that after accidentally running the tool under their Home Directory, all of their personal SSH keys, password manager vault, photos, and videos were packaged and uploaded to xAI’s servers.
(Background: Musk wants Tesla employees to use Grok to save money, even though he admits Claude Fable is stronger)
(Additional context: SpaceXAI, Musk’s company, has officially launched the most powerful model “Grok 4.5”! Teaming up with Cursor to heavily attack AI coding agents, with perfect integration into Office productivity software)

Table of contents

Toggle

• The entire home directory was packaged and uploaded
• Community warned early: .env files are also not spared
• Sounding a security alarm for AI development tools

With the widespread adoption of AI-assisted development tools, the boundary between code and personal privacy data is facing unprecedented challenges. Recently, Grok Build, a development tool launched by xAI under Elon Musk, sparked a major uproar in the community.

The entire home directory was packaged and uploaded

On July 13, 2026, Taipei time, a developer named A Green Being (@a_green_being) posted a chilling screenshot of logs (Logs) on the X platform. In a panic, they said: “Grok has uploaded my entire user directory to xAI’s servers. It contains my SSH keys, my password manager database, my personal documents, photos, videos—everything…”

According to the screenshot of the unified.json log they shared, the system indeed triggered a large number of repo_state.upload.start requests, and the upload path directly pointed to the user’s home directory (/home/usuario). This means Grok Build, without explicit selection and exclusion, transmitted all local files containing extremely sensitive information to a cloud endpoint.

Okay, grok has uploaded my entire user directory to xAI's servers. It contains my SSH keys, my password manager database, my documents, photos, videos, everything… pic.twitter.com/s8eSs9DGzF

— A Green Being (@a_green_being) July 13, 2026

Community warned early: .env files are also not spared

In fact, this privacy disaster is not an isolated case. A Green Being’s post was in response to another well-known community member, CyberSatoshi (@XBToshi), who had previously issued a security warning. CyberSatoshi had said that Grok Build would quietly upload the entire repository, Git history records, and even hidden .env files containing sensitive information such as API keys to the Google Cloud platform (GCP), which very likely could be used for subsequent model training.

CyberSatoshi then urged all developers to check the Grok log files on their local machines as soon as possible to confirm whether data was leaked. And facing A Green Being’s situation where their private data was fully “stripped,” he replied helplessly: “You’re worse than me—you ran Grok directly in your home directory.”

Sounding a security alarm for AI development tools

This incident highlights potential risks in today’s AI development tools regarding permission control and data collection. Since AI systems typically need a lot of context to understand programming logic, if a tool’s underlying design lacks proper file exclusion mechanisms (such as automatically ignoring .gitignore or system hidden files), it can easily turn into a security vulnerability on a disaster scale.

At present, cybersecurity experts strongly recommend that if developers use Grok Build or similar AI-assist tools, they should absolutely avoid running them under the root directory or home directory. The best protective practice is to limit these tools to a separate sandbox virtual environment, Docker containers, or run them under a dedicated restricted user account. In addition, users should constantly monitor abnormal data upload behavior via terminal commands (such as cat ~/.grok/logs/unified.json | grep repo_state.upload) to prevent personal digital assets from becoming free training fodder for AI.