ONNX disclosed CVE-2026-115922, a medium-severity denial-of-service vulnerability affecting all versions through 1.21.0, patched in version 1.22.01. The flaw allowed a crafted model as small as 66 bytes to trigger approximately 700 MB of memory allocation by declaring 700 million elements backed by a single packed byte. The vulnerable _unpack_2bit() and _unpack_4bit() code paths resized arrays to attacker-controlled dimensions without verifying the packed payload contained enough logical elements. Affected tensor types include UINT2, INT2, UINT4, INT4, and FLOAT4E2M1 stored in raw_data or packed int32_data.
ONNX Patches Denial-of-Service Flaw in Sub-Byte Tensor Handling
CVE-2026-115922 allows a 66-byte ONNX model to allocate roughly 700 MB via unchecked sub-byte tensor expansion; version 1.22.0 patches the flaw.
1[medium] onnx: ONNX: Denial of service via unchecked expansion of packed sub-byte tensors in ONNX Python APIsgithub.comHigh10·28 21:16Z
Was this useful?
The day’s vector, in your inbox.
Vector Wire Daily: one dispatch each publishing day.