AI Stack Alert Medium CVE-2026-115921 For AI Eng, AppSec, SOC / IR AI-assisted
ONNX <=1.21.0: Null pointer and unchecked narrowing in Scan shape inference (CVE-2026-115921)
ONNX versions 1.4.0 through 1.21.0 contain a flaw in Scan shape inference where the required num_scan_inputs attribute is not validated. A missing attribute causes a null pointer dereference, and a negative value causes a std::length_error during vector allocation. Both can terminate the process; the fix is in ONNX 1.22.0.
Why it matters
Any application that runs ONNX shape inference on untrusted model files can be crashed or denied service. The negative-value trigger passes onnx.checker.check_model(), so model validation alone does not catch it.
Do this
- Upgrade ONNX to 1.22.0 or later in all environments that load or process models.
- Inventory services that call onnx shape inference or model loading on untrusted or user-supplied models, and prioritize them for upgrade.
- Until patched, reject or sandbox untrusted ONNX models before shape inference, and run inference in an isolated worker process so a crash does not take down the main service.
- If the application exposes ONNX model processing over a network, score the issue with Modified Attack Vector set to Network, as the advisory recommends.
Exploit Clock
Exploitation unlikely (for now)
- EPSS 0.13%
Signals
- NoCISA KEV
- 6.2CVSS
- 0.13%EPSS
- 28/100Priority
- 90%Confidence
Evidence
“If the attribute was missing, `ctx.getAttribute("num_scan_inputs")` returned null and was dereferenced.”
“The negative-value variant passed `onnx.checker.check_model()` before reaching shape inference.”
“Upgrade to ONNX 1.22.0 or later.”
Sources
Affected / related
onnx, Scan operator shape inference
Published 2026-10-28 21:16 UTC · updated 2026-11-08 02:33 UTC · AI-assisted summary, checked against quoted sources · How we score