CVE-2026-34446: ONNX hardlink path traversal leaks

CISO Take

ONNX's onnx.load() bypasses path traversal protections for hardlinks, allowing a malicious model file to read arbitrary files accessible to the loading process. This is a supply chain vector: any team loading ONNX models from untrusted sources is exposed. Patch to 1.21.0 immediately and enforce trusted-source-only model loading policies.

What is the risk?

Medium risk overall, but elevated in AI/ML supply chain contexts. CVSS 4.7 reflects local-only, high-complexity, user-interaction-required conditions. Real-world risk increases significantly when ONNX models are sourced from public repositories or shared storage—common in ML workflows. The high confidentiality impact (API keys, credentials, model weights on the same filesystem) warrants prompt attention despite the medium CVSS.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
ONNX	pip	<= 1.20.1	`1.21.0`
21.1K OpenSSF 8.8 1.2K dependents Pushed 6d ago 91% patched ~44d to patch Full package profile →

Do you use ONNX? You're affected.

How severe is it?

CVSS 3.1

4.7 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 7% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

No known exploitation

Sophistication

Moderate

What is the attack surface?

AV Local

AC High

PR None

UI Required

S Unchanged

C High

I None

A None

What should I do?

6 steps

Patch: Upgrade onnx to >= 1.21.0 immediately (pip install --upgrade onnx).
Audit: Identify all systems calling onnx.load() against externally-sourced models.
Policy: Enforce cryptographic signing or hash verification of ONNX model files before loading.
Isolation: Run model loading in sandboxed environments with minimal filesystem access (container or VM with restricted mounts).
Detection: Monitor for unexpected file reads during model loading via auditd or eBPF.
Interim workaround if patching is blocked: validate all model sources are fully trusted and restrict filesystem permissions for the loading process user.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

Technical Impact partial

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

How is it classified?

Supply Chain Data Extraction Framework Model AML.T0002.001 - Models AML.T0010.001 - AI Software AML.T0011.000 - Unsafe AI Artifacts AML.T0037 - Data from Local System

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.10.1 - Supplier relationships and AI supply chain

NIST AI RMF

GOVERN 6.2 - Policies and procedures for AI risk across the supply chain

OWASP LLM Top 10

LLM03 - Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2026-34446?

ONNX's onnx.load() bypasses path traversal protections for hardlinks, allowing a malicious model file to read arbitrary files accessible to the loading process. This is a supply chain vector: any team loading ONNX models from untrusted sources is exposed. Patch to 1.21.0 immediately and enforce trusted-source-only model loading policies.

Is CVE-2026-34446 actively exploited?

No confirmed active exploitation of CVE-2026-34446 has been reported, but organizations should still patch proactively.

How to fix CVE-2026-34446?

1. Patch: Upgrade onnx to >= 1.21.0 immediately (pip install --upgrade onnx). 2. Audit: Identify all systems calling onnx.load() against externally-sourced models. 3. Policy: Enforce cryptographic signing or hash verification of ONNX model files before loading. 4. Isolation: Run model loading in sandboxed environments with minimal filesystem access (container or VM with restricted mounts). 5. Detection: Monitor for unexpected file reads during model loading via auditd or eBPF. 6. Interim workaround if patching is blocked: validate all model sources are fully trusted and restrict filesystem permissions for the loading process user.

What systems are affected by CVE-2026-34446?

This vulnerability affects the following AI/ML architecture patterns: model serving, training pipelines, MLOps/CI-CD pipelines, model conversion workflows, research/notebook environments.

What is the CVSS score for CVE-2026-34446?

CVE-2026-34446 has a CVSS v3.1 base score of 4.7 (MEDIUM). The EPSS exploitation probability is 0.18%.

What is the AI security impact?

Affected AI Architectures

model servingtraining pipelinesMLOps/CI-CD pipelinesmodel conversion workflowsresearch/notebook environments

MITRE ATLAS Techniques

AML.T0002.001 Models

AML.T0010.001 AI Software

AML.T0011.000 Unsafe AI Artifacts

AML.T0037 Data from Local System

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.10.1

NIST AI RMF: GOVERN 6.2

OWASP LLM Top 10: LLM03

What are the technical details?

Original Advisory

Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. Prior to version 1.21.0, there is an issue in onnx.load, the code checks for symlinks to prevent path traversal, but completely misses hardlinks because a hardlink looks exactly like a regular file on the filesystem. This issue has been patched in version 1.21.0.

Exploitation Scenario

An adversary creates a malicious ONNX model file containing hardlinks pointing to sensitive files on the target system (e.g., ~/.ssh/id_rsa, .env files with API keys, database credentials, or proprietary model weights). They publish this model to a public hub such as Hugging Face or GitHub, or submit it through a model upload feature on a platform. A data scientist or automated CI/CD pipeline calls onnx.load('malicious_model.onnx')—the hardlink bypasses symlink detection entirely, and the target file contents become accessible through the model structure. In CI/CD pipelines with broad filesystem access, this could expose build secrets to an adversary who controls the model artifact.

Weaknesses (CWE)

CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal') Primary CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal') Primary CWE-61 UNIX Symbolic Link (Symlink) Following Primary CWE-61 UNIX Symbolic Link (Symlink) Following 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.