Null Pointer Dereference in ONNX Model Conversion

Compliance Impact

The provided information does not specify any direct impact of CVE-2026-44512 on compliance with common standards and regulations such as GDPR or HIPAA.

Useful 0 Not Useful 0

Executive Summary

CVE-2026-44512 is a null pointer dereference vulnerability in the ONNX (Open Neural Network Exchange) framework's version converter adapters. It occurs in the function Upsample_6_7::adapt_upsample_6_7() when processing a model that contains an Upsample node with zero inputs. The adapter attempts to access the first input without verifying that any inputs exist, leading to a crash such as a segmentation fault or abort.

This issue affects ONNX versions from 1.9.0 up to but not including 1.22.0 and can cause an unrecoverable denial of service when converting untrusted models with malformed Upsample nodes.

The root cause is missing bounds checking on node inputs in several version converter adapters, which was fixed by adding explicit precondition checks to ensure the minimum required inputs or outputs exist before accessing them.

Useful 0 Not Useful 0

Impact Analysis

This vulnerability can cause an unrecoverable denial of service (DoS) in applications that use ONNX's version converter to process untrusted or malformed machine learning models.

Specifically, if a model contains an Upsample node with zero inputs, the version conversion process may crash due to a null pointer dereference, resulting in a segmentation fault or abort.

This can disrupt workflows, cause application crashes, and potentially lead to service downtime if the affected ONNX versions are used in production environments.

Useful 0 Not Useful 0

Detection Guidance

This vulnerability occurs when the ONNX version converter processes a model containing an Upsample node with zero inputs, causing a null pointer dereference and crash during version conversion.

To detect this vulnerability, you can attempt to convert untrusted ONNX models using the vulnerable versions (1.9.0 up to before 1.22.0) with the function onnx.version_converter.convert_version(). If the process crashes or aborts, it indicates the presence of the vulnerability.

Specifically, testing with a proof-of-concept model that includes an Upsample node with zero inputs can trigger the issue.

There are no explicit network or system commands provided in the resources to detect this vulnerability automatically.

Useful 0 Not Useful 0

Mitigation Strategies

The primary mitigation is to upgrade the ONNX library to version 1.22.0 or later, where this vulnerability has been fixed by adding explicit precondition checks to prevent null pointer dereferences.

If upgrading is not immediately possible, avoid processing untrusted ONNX models that may contain malformed Upsample nodes with zero inputs using the vulnerable version converter.

Additionally, review and validate ONNX models before conversion to ensure they do not contain nodes with invalid input counts.

Useful 0 Not Useful 0