Open Neural Network Exchange Vulnerability in ONNX Prior to Version 1.22.0
CVE-2026-115919

6.2MEDIUM

Key Information:

Vendor

Onnx

Status
Vendor
CVE Published:
26 October 2026

What is CVE-2026-115919?

The Open Neural Network Exchange (ONNX) prior to version 1.22.0 is vulnerable due to a flaw in the RNNShapeInference_opset1_to_6() function. This function improperly processes malformed Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) nodes from opsets 1 through 6, as well as Gated Recurrent Unit (GRU) nodes from opsets 3 through 6, without verifying the rank of the first input. Consequently, this can lead to out-of-bounds memory access on an undersized protobuf shape, which might cause applications to crash when processing untrusted models using onnx.shape_inference.infer_shapes(). The issue is resolved in version 1.22.0.

Affected Version(s)

onnx < 1.22.0

References

CVSS V3.1

Score:
6.2
Severity:
MEDIUM
Confidentiality:
None
Integrity:
None
Availability:
None
Attack Vector:
Local
Attack Complexity:
Low
Privileges Required:
None
User Interaction:
None
Scope:
Unchanged

Timeline

  • Vulnerability published

  • Vulnerability Reserved

.