CVE-2026-39987 Marimo Pre-auth RCE Research

Overview

• This project focuses on CVE-2026-39987, a pre-authentication Remote Code Execution vulnerability in Marimo related to the terminal WebSocket endpoint /terminal/ws.

• Marimo is an open-source reactive Python notebook that allows users to write and execute Python code through a web interface. Because notebook platforms can execute code on the server side, privileged features such as terminal access must be protected by proper authentication and access control.

• In vulnerable Marimo versions, authentication was not properly enforced on the terminal WebSocket endpoint. As a result, an unauthenticated user could potentially access terminal functionality and execute commands in the environment where Marimo is running.

• The goal of this project is to reproduce the vulnerability in a local Docker lab, analyze the root cause, compare the vulnerable and patched versions, and verify that the fix prevents the issue.

CVE Information

• CVE ID: CVE-2026-39987
• Product: Marimo
• Vulnerability type: Missing Authentication for Critical Function
• CWE: CWE-306
• Affected versions: Marimo < 0.23.0
• Fixed version: Marimo 0.23.0+
• Affected endpoint: /terminal/ws
• Impact: Pre-authenticated Remote Code Execution
• Severity: Critical
• CVSS v4.0: 9.3
• CVSS v3.1: 9.8

Project Goals

• Understand Marimo and the vulnerability context
• Build a vulnerable local Docker lab
• Reproduce the vulnerability safely in localhost
• Analyze the root cause in /terminal/ws
• Perform patch diffing between vulnerable and fixed versions
• Verify the fix on patched Marimo
• Write mitigation and lessons learned

Lab Environments

• OS: Windows 11
• Docker version: 29.0.0
• Docker Compose version: v5.5.1
• Python version: 3.11.9
• Vulnerable Marimo version: TODO
• Patched Marimo version: TODO

Project Status

• Research CVE
• Build vulnerable lab
• Build patched lab
• Reproduce PoC
• Root cause analysis
• Patch diff
• Verify fix
• Final report

Disclaimer

This project is for educational purposes only. All testing will be performed in a local Docker lab environment.