JupyterLab Command Execution via Crafted HTML Content

A vulnerability exists in JupyterLab and Notebook where specially crafted HTML content can be embedded within a notebook or Markdown file. This content leverages the data-commandlinker-command and data-commandlinker-args attributes, which are improperly sanitized. When a user opens the malicious notebook or Markdown file and clicks on a deceptively crafted button, it triggers the execution of arbitrary JupyterLab commands without further user interaction or code submission. This can lead to arbitrary code execution, file deletion, or denial of service. The vulnerability affects JupyterLab versions 4.5.6 and earlier, as well as Notebook versions 7.0.0 through 7.5.5. A patch is available in JupyterLab 4.5.7 to address this issue. This vulnerability poses a significant risk as it only requires a single click from the user to initiate malicious actions.

Attack Chain

1. Attacker crafts a malicious notebook or Markdown file containing an HTML cell output.
2. The HTML cell includes a <button> element with data-commandlinker-command and data-commandlinker-args attributes set to a malicious command.
3. The attacker distributes the malicious file via email, GitHub, or a Binder link.
4. Victim opens the file in JupyterLab or Notebook.
5. The malicious HTML is rendered in the output area, displaying a deceptive button visually indistinguishable from legitimate widgets.
6. The victim clicks on the button.
7. CommandLinker captures the click event on document.body.
8. The CommandLinker executes the command specified in the data-commandlinker-command attribute, leading to arbitrary code execution, file deletion, or other malicious actions.

Impact

Successful exploitation of this vulnerability can lead to arbitrary code execution within the context of the JupyterLab server. This could allow an attacker to delete files, potentially causing unrecoverable data loss. In multi-tenant environments, this could lead to denial-of-service by exhausting server resources. In certain browser configurations (Chromium-based), attackers can potentially gain full terminal access via multi-click attacks combined with clipboard access. The affected products are JupyterLab versions up to 4.5.6 and Notebook versions 7.0.0 through 7.5.5.

Recommendation

• Upgrade to JupyterLab version 4.5.7 or later to patch CVE-2026-42557.
• For downstream applications, disable the CommandLinker during initialization as described in the advisory.
• Implement the hardening steps by setting "allowCommandLinker": false in the overrides.json file.
• Educate users about the risks of opening notebooks and Markdown files from untrusted sources to prevent T1204.002 (User Execution).