Grok Data Exposure Checker

A standalone tool that detects whether the Grok Build CLI uploaded your data to xAI's servers without your knowledge.

The Problem

Grok Build CLI (grok) silently packages and uploads workspace contents to xAI's Google Cloud Storage (gs://grok-code-session-traces). When launched from a home directory or broad path, it can attempt to upload your entire home directory — including SSH keys, API tokens, cloud credentials, and personal files.

This tool reads Grok's own log files to determine:

• Whether any uploads occurred
• What data was in the uploaded paths
• Whether sensitive files were exposed
• What you should do about it

Quick Start

python3 scripts/grok_exposure_check.py

That's it. The tool generates a styled HTML report and opens it in your browser.

Requirements: Python 3.8+. No pip installs. No dependencies. No network access.

What You'll See

The report is a styled HTML page with a visual traffic-light status:

The traffic-light status:

• GREEN — No upload events found. You're clear.
• YELLOW — Upload was attempted but outcome is uncertain. Rotate credentials as a precaution.
• RED — Confirmed data upload to xAI servers. Rotate credentials immediately.

The report includes:

• Every upload event found in the logs with timestamps and destinations
• Sensitive files that existed in the uploaded paths, sorted by severity
• Telemetry history (when uploads were enabled/disabled)
• Data retention opt-out status
• Specific recommended actions based on your exposure level

How It Works

The tool reads ~/.grok/logs/unified.jsonl (Grok's unified log) and looks for:

Log Event	Meaning
repo_state.upload.start	Grok began packaging a directory for upload
repo_state.upload.enqueued	Upload confirmed with a GCS path (data reached xAI)
trace.upload.decision	Whether telemetry/uploads were enabled per session
repo_state.upload.skip	Upload was explicitly skipped (safe)

It then cross-references the upload paths against a database of known sensitive file locations (SSH keys, cloud credentials, environment files, etc.) to determine what was at risk.

Sensitive File Detection

Files are classified by severity:

• CRITICAL — Credentials allowing impersonation (SSH private keys, .env files, OAuth tokens, AWS credentials, database files)
• HIGH — Sensitive data worth protecting (API keys in config files, package manager tokens)
• MEDIUM — Config that may contain secrets (.gitconfig, SSH authorized_keys)
• LOW — Minimal risk metadata (SSH known_hosts)

Files checked include:

• SSH keys and configs (.ssh/id_*, authorized_keys, config)
• Cloud credentials (.aws/credentials, .config/gcloud/, .azure/, .kube/config)
• Environment files (.env, .env.local, .env.production)
• Auth tokens (.grok/auth.json, .npmrc, .pypirc, .netrc)
• Docker, Git, GPG, crypto wallets
• Hermes Agent config (.hermes/config.yaml, .hermes/.env)

Cross-Platform Support

Platform	Grok Home Location
Linux	~/.grok
macOS	~/.grok
Windows	%USERPROFILE%\.grok
Custom	Set GROK_HOME environment variable

Using With AI Agents

This tool works with any AI coding agent — Hermes, Cursor, Claude Code, Codex, Windsurf, Cline, etc.

When you ask an agent "check if Grok exposed my data," it should run the script and present the HTML report. See AGENTS.md for detailed agent instructions.

Privacy

This tool:

• Reads files only under ~/.grok/ and checks file existence at known paths
• Does NOT transmit any data over the network
• Does NOT read file contents (only checks if files exist)
• Does NOT modify any files

Limitations

• Only checks local logs — server-side data may differ
• Cannot determine with certainty if a before_codebase upload completed when no .enqueued event exists
• Cannot delete data from xAI's servers
• Sensitive file detection uses known patterns; novel file names may be missed

Background

The Grok Build CLI is xAI's autonomous coding agent (similar to Claude Code or Codex CLI). It launched in May 2026 as part of SuperGrok Heavy. In July 2026, wire-level analysis revealed that the CLI uploads workspace contents to xAI's cloud as part of its normal operation — including cases where users launched it from their home directory.

Key findings from the analysis:

• Uploads go to gs://grok-code-session-traces (private GCS bucket)
• The upload route is: your machine -> cli-chat-proxy.grok.com -> GCS
• The backend code service is at code.grok.com
• A fix in v0.2.90 addressed folder trust for home directories that are git repos
• Telemetry (and thus uploads) can be enabled/disabled server-side by xAI

File Structure

grok-exposure-checker/
├── README.md                        This file
├── AGENTS.md                        Instructions for AI agents
├── LICENSE                          MIT
├── scripts/
│   └── grok_exposure_check.py       The standalone tool
└── docs/
    └── example-report.html          Example sanitized report

License

MIT