Out-of-bounds read in ONNX RNN, GRU and LSTM shape inference
Key facts
- Attack vector
- Local
- Privileges required
- None
- User interaction
- None
- Attack complexity
- Low
- Impact type
- Denial of service
- Confidentiality · integrity · availability
- None · None · High
- Fix status
- Partial Red Hat: 1 without a fix
Attack profile from CVSS 3.1 · CNA. Fix status from CNA, GitHub and Red Hat. Impact type from CWE-125 and the CVSS impact.
Summary
ONNX versions before 1.22.0 are affected. The RNN, GRU and LSTM shape inference functions for opsets 1 through 6 do not check that the first input has rank three, so they read outside the intended protobuf shape field (CWE-125). An attacker who supplies an untrusted model that passes onnx.checker.check_model() can crash an application that calls onnx.shape_inference.infer_shapes() on it.
Written with Claude Haiku 5.5 from the sources below · 2026-11-06
Description from the CNA
Open Neural Network Exchange (ONNX) is an open standard for machine learning interoperability. Prior to 1.22.0, RNNShapeInference_opset1_to_6() processes malformed RNN and LSTM nodes from opsets 1 through 6 and GRU nodes from opsets 3 through 6 without verifying that the first input has rank three. The function accesses dim(0) and dim(1) on an undersized protobuf shape, which can read outside the intended repeated field and terminate an application when onnx.shape_inference.infer_shapes() processes an untrusted model that can pass onnx.checker.check_model(). This issue is fixed in version 1.22.0.
Why it matters now
SSVC decision points
CISA ADP · SSVC 2.0.3
- Exploitation
- None
- Automatable
- No
- Technical impact
- Partial
The decision (Track, Attend, Act) also depends on your mission and well-being impact; see CISA's SSVC guide.
Affected products
| Vendor | Product | Affected versions | Fixed in | Status | Source |
|---|---|---|---|---|---|
| onnx | onnx | before 1.22.0 | 1.22.0 | Patch available | CNA |
| — | onnx (PyPI) | before 1.22.0 | 1.22.0 | Patch available | GHSA-48m4-24cq-37qf |
- onnx Vendoronnx Affected before 1.22.0 Fixed in 1.22.0 Patch available CNA
- onnx (PyPI) Vendor— Affected before 1.22.0 Fixed in 1.22.0 Patch available GHSA-48m4-24cq-37qf
Remediation
Upgrade ONNX to version 1.22.0 or later for both the onnx product and the PyPI package. Red Hat lists no fix available for Red Hat OpenShift AI (RHOAI).
Written with Claude Haiku 5.5 from the sources below · 2026-11-06
- Fix · CNA onnx (onnx) Fixed in 1.22.0
- Fix · GHSA-48m4-24cq-37qf onnx (PyPI) Fixed in 1.22.0 Advisory →
Vendor assessments
Red Hat
Red Hat rating: Moderate. Red Hat's page for this CVE →
| Product | Status | Plan or fix |
|---|---|---|
| Red Hat OpenShift AI (RHOAI) | Affected | No fix yet: Fix deferred |
Timeline
- 2026-10-26 Threatist record created from CVE-2026-115919
- 2026-10-26 CISA SSVC decision points changed
- 2026-10-26 Priority changed: MEDIUM → LOW
- 2026-10-27 Affected versions changed · Red Hat
- 2026-10-27 CVSS score added · Red Hat
- 2026-10-28 Alias added: GHSA-48m4-24cq-37qf
- 2026-10-28 Fix available: PyPI/[email protected] · GHSA-48m4-24cq-37qf
- 2026-10-28 CVSS score added · GHSA-48m4-24cq-37qf
- 2026-10-28 Weakness changed · GHSA-48m4-24cq-37qf
- 2026-10-28 6 new references · GHSA-48m4-24cq-37qf
References
More onnx vulnerabilities
- TI-2026-22763 ONNX Untrusted Model Repository Warnings Suppressed by silent=True in onnx.hub.load() — Silent Supply-Chain Attack CVSS 3.1CRITICAL 9.1
- TI-2026-21896 ONNX: Path Traversal via Symlink CVSS 4.0HIGH 8.7
- TI-2024-07209 Arbitrary File Overwrite in onnx/onnx CVSS 3.1CRITICAL 9.1
- TI-2024-04814 Arbitrary File Overwrite in download_model_with_test_data in onnx/onnx CVSS 3.1HIGH 8.8
- TI-2026-36338 ONNX symlink-following and path-traversal arbitrary file write CVSS 4.0MEDIUM 6.8
- TI-2026-32442 ONNX: Null Pointer Dereference in Upsample Version Converter Adapter (Zero Inputs) CVSS 3.1MEDIUM 5.5
Sources
- CVE-2026-115919 · CVE Program
- CVE-2026-115919 · NVD
- FIRST EPSS
- GHSA-48m4-24cq-37qf · GitHub Advisory Database · CC BY 4.0
- CVE-2026-115919 · Red Hat security data · CC BY 4.0
This product uses the NVD API but is not endorsed or certified by the NVD. Sources and licences