AI Threat Alert
CVE-2026-39378: nbconvert: path traversal exfiltrates files via HTML export
GHSA-7jqv-fw35-gmx9 MEDIUMJupyter's nbconvert contains a path traversal vulnerability that allows a malicious notebook to read arbitrary files from the conversion host and silently embed them as base64-encoded data URIs in the exported HTML — effectively turning your report generation pipeline into an exfiltration channel. With 2,931 downstream dependents and EPSS at the 89th percentile, exposure across ML engineering teams is substantial, particularly in CI/CD automation and collaborative platforms that process external notebooks. The exploit is non-default (requires HTMLExporter.embed_images=True to be explicitly enabled) and technically trivial once that condition is met, making it an attractive low-effort pivot for supply chain attackers who can influence which notebooks your pipelines consume. Patch to nbconvert 7.17.1 immediately and audit all pipeline configurations for explicit embed_images=True settings.
What is the risk?
CVSS 6.5 (Medium) understates real-world risk in ML pipeline contexts. The confidentiality sub-score is HIGH (C:H) — sensitive files including SSH keys, API tokens, and model configurations are fully readable by any notebook processed by an affected pipeline. The non-default flag requirement narrows exposure but any team that has enabled embed_images=True is fully compromised by whatever notebooks they ingest. OpenSSF Scorecard of 4.9/10 and 9 prior CVEs in the same package signal a persistently weak security posture. SSVC is TRACK with no active exploitation confirmed, but the trivial exploitation difficulty warrants immediate patching rather than deferral.
How does the attack unfold?
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| Jupyter Notebook | pip | >= 6.5.0, < 7.17.1 | 7.17.1 |
Do you use Jupyter Notebook? You're affected.
How severe is it?
What is the attack surface?
What should I do?
5 steps-
PATCH
Upgrade nbconvert to 7.17.1 across all environments (dev, staging, production).
-
AUDIT
Search all nbconvert configurations, CI/CD scripts, and Makefile targets for 'embed_images=True' or '--embed-images' flags — this setting is False by default but may have been enabled for self-contained report generation.
-
SANDBOX
Run notebook conversion in isolated containers with read-only filesystem mounts limited to notebook working directories.
-
SUPPLY CHAIN
Validate the provenance of notebooks processed by automated pipelines; treat external notebooks as untrusted input.
-
DETECT
Monitor HTML export jobs for unexpectedly large output files (embedded file contents inflate HTML size significantly).
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-2026-39378?
Jupyter's nbconvert contains a path traversal vulnerability that allows a malicious notebook to read arbitrary files from the conversion host and silently embed them as base64-encoded data URIs in the exported HTML — effectively turning your report generation pipeline into an exfiltration channel. With 2,931 downstream dependents and EPSS at the 89th percentile, exposure across ML engineering teams is substantial, particularly in CI/CD automation and collaborative platforms that process external notebooks. The exploit is non-default (requires HTMLExporter.embed_images=True to be explicitly enabled) and technically trivial once that condition is met, making it an attractive low-effort pivot for supply chain attackers who can influence which notebooks your pipelines consume. Patch to nbconvert 7.17.1 immediately and audit all pipeline configurations for explicit embed_images=True settings.
Is CVE-2026-39378 actively exploited?
No confirmed active exploitation of CVE-2026-39378 has been reported, but organizations should still patch proactively.
How to fix CVE-2026-39378?
1. PATCH: Upgrade nbconvert to 7.17.1 across all environments (dev, staging, production). 2. AUDIT: Search all nbconvert configurations, CI/CD scripts, and Makefile targets for 'embed_images=True' or '--embed-images' flags — this setting is False by default but may have been enabled for self-contained report generation. 3. SANDBOX: Run notebook conversion in isolated containers with read-only filesystem mounts limited to notebook working directories. 4. SUPPLY CHAIN: Validate the provenance of notebooks processed by automated pipelines; treat external notebooks as untrusted input. 5. DETECT: Monitor HTML export jobs for unexpectedly large output files (embedded file contents inflate HTML size significantly).
What systems are affected by CVE-2026-39378?
This vulnerability affects the following AI/ML architecture patterns: Jupyter notebook pipelines, ML training pipelines, CI/CD automation, Automated report generation, Data science workflows.
What is the CVSS score for CVE-2026-39378?
CVE-2026-39378 has a CVSS v3.1 base score of 6.5 (MEDIUM). The EPSS exploitation probability is 0.04%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0010.001 AI Software AML.T0011.000 Unsafe AI Artifacts AML.T0025 Exfiltration via Cyber Means AML.T0037 Data from Local System Compliance Controls Affected
What are the technical details?
Original Advisory
## Summary When `HTMLExporter.embed_images=True`, nbconvert's markdown renderer allows arbitrary file read via path traversal in image references. A malicious notebook can exfiltrate sensitive files from the conversion host by embedding them as base64 data URIs in the output HTML. ## Patches Upgrade to nbconvert 7.17.1 ## Workarounds Do not enable `HTMLExporter.embed_images` (it is not enabled by default).
Exploitation Scenario
A red teamer submits a crafted Jupyter notebook to an internal ML platform that auto-converts notebooks to HTML for stakeholder reporting, with embed_images=True enabled for self-contained reports. The notebook includes image display cells referencing paths like '../../../../home/ubuntu/.aws/credentials' and '../../../../etc/passwd' via relative traversal. When the CI/CD pipeline runs nbconvert, these files are read from the host and embedded as base64 data URIs in the output HTML report. The attacker retrieves the published report, decodes the data URIs, and obtains AWS credentials and system user information — all without any anomalous network activity or authentication events.
Weaknesses (CWE)
CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')
Primary
CWE-23 Relative Path Traversal
Primary
CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted directory.
- [Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
- [Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:N/A:N References
Timeline
Related Vulnerabilities
CVE-2026-42266 8.8 JupyterLab: Extension allow-list bypass enables privesc
Same package: notebook CVE-2026-5422 8.1 jupyter-server: path traversal exposes sibling dir files
Same package: notebook CVE-2018-8768 7.8 Jupyter Notebook: XSS via malicious .ipynb file
Same package: notebook CVE-2026-35397 7.1 Jupyter Server: path traversal leaks sibling directories
Same package: notebook CVE-2026-39377 6.5 nbconvert: path traversal enables arbitrary file write
Same package: notebook