CVE-2026-115921

Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. From 1.4.0 until 1.22.0, ScanInferenceFunction, ScanInferenceFunction opset8, and ScanInferenceFunction opset9 do not safely validate the required num scan inputs attribute. A missing attribute is dereferenced as a null pointer, while a negative value is converted to size t and can reach an exceptional oversized allocation path. An untrusted model can therefore terminate or disrupt shape inference, including a negative-value model that can pass onnx.checker.check model(). This issue is fixed in version 1.22.0.