An out‑of‑bounds read occurs in ONNX ConvTranspose shape inference when the weight tensor has fewer than three dimensions and no kernel_shape attribute is supplied. The error is triggered during the calculation of an empty effective kernel shape and results in a process crash. Because the crash is caused by untrusted model files that have successfully passed onnx.checker.check_model(), the vulnerability presents a denial‑of‑service risk; there is no evidence of confidentiality or integrity compromise. The weakness is identified as CWE‑125.

The flaw is present in the ONNX library for all releases from version 1.2.1 up to, but not including, 1.22.0. It affects the convTransposeShapeInference_opset11() function and the current inference path used by applications that perform shape inference on incoming model files.

The CVSS score of 6.2 indicates a moderate severity. Since the EPSS score is not available and the vulnerability is not listed in the CISA KEV catalog, the probability of exploitation in the wild is currently uncertain. An attacker would need to supply a malicious ONNX model that can pass the ONNX checker and is processed by shape inference. The impact is limited to service interruption for the application that loads the model; it does not allow arbitrary code execution or data exfiltration.