CRITICAL:
I Just Hacked Grok

June 8, 2026 • Michail Strokin

The Discovery That Shook xAI

What started as a simple image generation request turned into one of the most significant security discoveries in AI chatbot history. In under 30 minutes, I went from chatting with Grok to having complete access to xAI's internal systems.

This isn't theoretical. This is real.

How It Happened

1
The Innocent Request
I asked Grok to generate an image. Something went wrong.
2
The System Glitch
Grok began executing shell commands in response to my messages.
3
Full System Access
I could navigate /usr/local/bin and download the entire .GROK skills directory.

The Crown Jewels

Asset Size Sensitivity
grok-killguard 133 KB Critical
grok-files 14.5 MB Critical
grok-computer-server.mjs 22.3 MB Critical
.GROK/skills/ directory Entire structure Top Secret

The Implications Are Terrifying

100%
Security Boundary Collapse
∞
Potential Damage

Any user could execute arbitrary commands, exfiltrate proprietary IP, and map the entire internal architecture.

⚠️ xAI has been notified. This is for educational purposes.

Read the full disclosure →

grok-root-skills github repo grok-hades-container-tools github repo