Out-of-bounds Read Vulnerability in Open Neural Network Exchange by ONNX
CVE-2026-63632

3.3LOW

Key Information:

Vendor

Onnx

Status
Vendor
CVE Published:
18 August 2026

What is CVE-2026-63632?

The Open Neural Network Exchange (ONNX) has a vulnerability in its version converter that can lead to an out-of-bounds read. Specifically, in the method 'onnx.version_converter.convert_version()', if a Gemm node has input tensors with fewer than two dimensions, it accesses B_shape[1], A_shape[0], or A_shape[1] without performing a rank check. This flaw can result in a process crash during a downgrade from opset 7 to 6, affecting the stability of applications relying on ONNX for machine learning interoperability. The issue has been addressed and fixed in version 1.22.0.

Affected Version(s)

onnx >= 1.3.0, < 1.22.0

References

CVSS V3.1

Score:
3.3
Severity:
LOW
Confidentiality:
None
Integrity:
None
Availability:
None
Attack Vector:
Local
Attack Complexity:
Low
Privileges Required:
None
User Interaction:
Required
Scope:
Unchanged

Timeline

  • Vulnerability published

  • Vulnerability Reserved

.