NVIDIA TensorRT for Linux: integer overflow allows denial of service and data tampering
Impact
TensorRT has an integer overflow that NVIDIA says can lead to denial of service and data tampering. Inference services built on TensorRT that process outside input could crash or produce corrupted results. The record does not name the trigger or the affected code path, and it does not claim code execution.
Who can reach it
Network, no authentication, but user interaction is required (CVSS UI:R). This most likely means a user or service has to process attacker-supplied content with TensorRT. The record does not say exactly what.
What to do
Upgrade TensorRT to the fixed release named in NVIDIA bulletin 5921, then rebuild or redeploy the applications and inference containers that bundle it. The record does not give a fixed version.
References
Related entries
- DGX Spark: Hardcoded credentials in configCVE-2026-24218 · DGX SparkHigh
- AI Tensor Engine for ROCm (AITER) - MessageQueue.recv() in shm_broadcast.py: AITER's MessageQueue.recv() deserialisesCVE-2026-49121 · AI Tensor Engine for ROCm (AITER) - MessageQueue.recv() in shm_broadcast.pyHigh
- DGX H100 BMC (KVM daemon): Session token theft via timing side channelCVE-2023-25529 · DGX H100 BMC (KVM daemon)High
- DGX H100 BMC (KVM): Code execution + privescCVE-2023-25530 · DGX H100 BMC (KVM)High
- Triton Inference Server: RCE / privesc via input-validation failureCVE-2025-23268 · Triton Inference ServerHigh
- Cumulus Linux / NVOS: Privesc to switch adminCVE-2025-33179 · Cumulus Linux / NVOSHigh
This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.