JupyterLab's PyPI Extension Manager fails to reject option-like values on uninstall requests, so an authenticated user can inject pip options that make the server read local files or fetch internal URLs, with the content leaking back through error messages. Real-world risk is low: it only applies when the PyPI Extension Manager is enabled and kernels and terminals are disabled or remote, because otherwise the user can already read files and make outbound requests directly. EPSS is 0.00206 (listed as top 90th percentile), the CVE is not in CISA KEV, there is no public exploit or Nuclei template, and SSVC says TRACK. The package has 1,892 dependents and 41 other CVEs, but this issue adds no code execution or availability impact. Upgrade to JupyterLab 4.5.11 or 4.6.4 in hardened multi-user deployments (JupyterHub with locked-down kernels), and disable the PyPI extension manager where it isn't needed.
What is the risk?
Medium (CVSS 5.4), and effectively lower in most deployments. It requires an authenticated low-privilege account with extension API access, and it matters only in locked-down setups where kernels and terminals are disabled or remote. Exploitation complexity is low, but the attacker cannot choose arbitrary file content or an arbitrary disclosed line. Confidentiality and integrity impacts are limited: a partial file or URL content leak through parse errors, and creation or corruption of a path via pip log text. There is no public exploit, no KEV listing and a low EPSS value. The main exposure is internal SSRF toward cloud metadata endpoints and internal services from the Jupyter server.
How does the attack unfold?
What systems are affected?
How severe is it?
What is the attack surface?
What should I do?
1 step1) Upgrade JupyterLab to 4.5.11 or 4.6.4 or later. 2) If patching is delayed, disable the PyPI Extension Manager (set the extension manager to a read-only or disabled mode, or use the
--LabApp.extension_manager=readonlystyle configuration) and restrict access to the extension API at the proxy or hub level. 3) Run the Jupyter server with a least-privilege account, so readable local files do not include secrets. 4) Block egress from the Jupyter host to cloud metadata endpoints (169.254.169.254) and internal admin networks, and enforce IMDSv2. 5) Detection: look in reverse proxy and Jupyter logs for POST requests to the extension API (/lab/api/extensions) whose uninstall name starts with-or--(for example--requirement,-r,--log), for pip parse errors echoed in responses, and for unexpected pip subprocess executions.
What does CISA's SSVC say?
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:
Frequently Asked Questions
What is CVE-2026-102904?
JupyterLab's PyPI Extension Manager fails to reject option-like values on uninstall requests, so an authenticated user can inject pip options that make the server read local files or fetch internal URLs, with the content leaking back through error messages. Real-world risk is low: it only applies when the PyPI Extension Manager is enabled and kernels and terminals are disabled or remote, because otherwise the user can already read files and make outbound requests directly. EPSS is 0.00206 (listed as top 90th percentile), the CVE is not in CISA KEV, there is no public exploit or Nuclei template, and SSVC says TRACK. The package has 1,892 dependents and 41 other CVEs, but this issue adds no code execution or availability impact. Upgrade to JupyterLab 4.5.11 or 4.6.4 in hardened multi-user deployments (JupyterHub with locked-down kernels), and disable the PyPI extension manager where it isn't needed.
Is CVE-2026-102904 actively exploited?
No confirmed active exploitation of CVE-2026-102904 has been reported, but organizations should still patch proactively.
How to fix CVE-2026-102904?
1) Upgrade JupyterLab to 4.5.11 or 4.6.4 or later. 2) If patching is delayed, disable the PyPI Extension Manager (set the extension manager to a read-only or disabled mode, or use the `--LabApp.extension_manager=readonly` style configuration) and restrict access to the extension API at the proxy or hub level. 3) Run the Jupyter server with a least-privilege account, so readable local files do not include secrets. 4) Block egress from the Jupyter host to cloud metadata endpoints (169.254.169.254) and internal admin networks, and enforce IMDSv2. 5) Detection: look in reverse proxy and Jupyter logs for POST requests to the extension API (`/lab/api/extensions`) whose uninstall name starts with `-` or `--` (for example `--requirement`, `-r`, `--log`), for pip parse errors echoed in responses, and for unexpected pip subprocess executions.
What systems are affected by CVE-2026-102904?
This vulnerability affects the following AI/ML architecture patterns: Notebook environments, Training pipelines, Multi-tenant JupyterHub deployments, ML development workbenches.
What is the CVSS score for CVE-2026-102904?
CVE-2026-102904 has a CVSS v3.1 base score of 5.4 (MEDIUM). The EPSS exploitation probability is 0.21%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0012Valid AccountsAML.T0037Data from Local SystemAML.T0049Exploit Public-Facing ApplicationCompliance Controls Affected
What are the technical details?
Original Advisory
JupyterLab is an extensible environment for interactive and reproducible computing, based on the Jupyter Notebook Architecture. From JupyterLab 4.0.0 until 4.5.11 and 4.6.4, the PyPI Extension Manager uninstall request reaches ExtensionHandler.post, which validates extension names for installation but passes uninstall names to PyPIExtensionManager.uninstall and python -m pip uninstall without rejecting option-like values. The security impact requires that the PyPI Extension Manager is enabled, the account can call the extension API, and kernels and terminals are disabled or delegated to remote hosts; otherwise the user can already read files and make outbound requests directly. An authenticated user with extension API access can supply a pip requirements option to make the server read a local file or fetch an internal URL, and reflected parse errors can return the first unparsable line or response content. A pip log option can also create or corrupt a chosen path with pip-generated log text, but the requester cannot select an arbitrary disclosed line or arbitrary file content, and the injection does not add code execution or availability impact beyond ordinary package removal. This issue is fixed in JupyterLab 4.5.11 and 4.6.4.
Exploitation Scenario
An attacker with a low-privilege account on a hardened shared JupyterLab service, where terminals and kernels are disabled, sends an uninstall request to the extension API with the name `-r /etc/<file>` or `-r http://169.254.169.254/latest/meta-data/`. The server runs `python -m pip uninstall` with the injected option, tries to parse the target as a requirements file, and returns the first unparsable line or response content in the error message. The attacker repeats this against internal URLs to map the network and collect cloud metadata, then uses a `--log` option to overwrite a chosen path with pip log text to corrupt a configuration file.
Weaknesses (CWE)
CWE-209Generation of Error Message Containing Sensitive InformationPrimaryCWE-209Generation of Error Message Containing Sensitive InformationPrimaryCWE-88Improper Neutralization of Argument Delimiters in a Command ('Argument Injection')PrimaryCWE-88Improper Neutralization of Argument Delimiters in a CommandPrimaryCWE-918Server-Side Request Forgery (SSRF)PrimaryCWE-918Server-Side Request ForgeryPrimaryCWE-209Generation of Error Message Containing Sensitive InformationCWE-88Improper Neutralization of Argument Delimiters in a CommandCWE-918Server-Side Request ForgeryCWE-209 — Generation of Error Message Containing Sensitive Information: The product generates an error message that includes sensitive information about its environment, users, or associated data.
- [Implementation] Ensure that error messages only contain minimal details that are useful to the intended audience and no one else. The messages need to strike the balance between being too cryptic (which can confuse users) or being too detailed (which may reveal more than intended). The messages should not reveal the methods that were used to determine the error. Attackers can use detailed information to refine or optimize their original attack, thereby increasing their chances of success. If errors must be captured in some detail, record them in log messages, but consider what could occur if the log messages can be viewed by attackers. Highly sensitive information such as passwords should never be saved to log files. Avoid inconsistent messaging that might accidentally tip off an attacker about internal state, such as whether a user account exists or not.
- [Implementation] Handle exceptions internally and do not display errors containing potentially sensitive information to a user.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:L/A:NReferences
- github.com/advisories/GHSA-3325-v43h-43rv
- github.com/jupyterlab/jupyterlab/commit/9e1951e57da499ca66ef5e0caf68e71ad2e3d1a5
- nvd.nist.gov/vuln/detail/CVE-2026-102904
- github.com/jupyterlab/jupyterlab/commit/a274a8276b9185d03efd4c3c20713d3137ac49e6
- github.com/jupyterlab/jupyterlab/commit/e277bc958e737130ac47b6c5078d08b29298828f
- github.com/jupyterlab/jupyterlab/releases/tag/v4.5.11
- github.com/jupyterlab/jupyterlab/releases/tag/v4.6.4
- github.com/jupyterlab/jupyterlab/security/advisories/GHSA-3325-v43h-43rv
Timeline
Related Vulnerabilities
CVE-2023-2557410.0JupyterHub LTI13: JWT forgery enables full auth bypass
Same package: jupyterCVE-2024-3970010.0JupyterLab extension-template: CI workflow RCE
Same package: jupyterCVE-2026-441809.8Jupyter Enterprise Gateway: root privilege bypass in Kubernetes
Same package: jupyterCVE-2026-235379.1Feast: unauth file write to RCE via /save-document
Same package: jupyterCVE-2026-545279.0jupyterlab-git: stored XSS escalates to full RCE
Same package: jupyter