Database/AI/ML frameworks & serving
NVIDIA DALI: stack-based buffer overflow in pipeline input handling allows code execution
Impact
A stack-based buffer overflow in DALI, which runs inside the training or inference process on the GPU node, is the most serious of the four issues in NVIDIA bulletin 2026/5900: NVIDIA lists code execution alongside disclosure, tampering and denial of service. Code execution lands with the privileges of the job that loaded the data, so on a multi-tenant cluster it means whoever controls the input controls that tenant's pod and any credentials, datasets or model weights mounted into it. The record does not describe an escape beyond that process, and the vector is local with user interaction, so this is a data-supply-chain problem rather than a remotely reachable one.
Who can reach it
Local, authenticated low-privileged user, with user interaction (CVSS AV:L/AC:L/PR:L/UI:R) - someone able to place a crafted input file where a DALI pipeline will decode it.
What to do
Upgrade DALI to the fixed version in NVIDIA bulletin 2026/5900, rebuild and redeploy the container images that pin it, and restart training and serving processes. Until then, treat pipeline inputs as untrusted and keep DALI jobs off shared credentials. The NVD record does not give a fixed version - confirm it in the bulletin.
References
Related entries
- NVIDIA DALI: out-of-bounds write in pipeline input handling corrupts memoryCVE-2026-65149 · NVIDIA DALI (data pipeline file/media decoding)High
- NVIDIA DALI: out-of-bounds reads while decoding pipeline input lead to disclosure and crashesCVE-2026-65147 · NVIDIA DALI (data pipeline file/media decoding)High
- JupyterHub: A user granted limited access can escalateCVE-2024-41942 · JupyterHubHigh
- Weaviate: Crafted entry name with an absolute pathCVE-2025-67818 · WeaviateHigh
- llama.cpp (`gguf_init_from_file`): Use of uninitialized heap variableCVE-2024-32878 · llama.cpp (`gguf_init_from_file`)High
- MLflow: basic-auth app leaves tracing and assessment endpoints without permission checksCVE-2025-15381 · MLflow Tracking Server (basic-auth app: tracing and assessment endpoints)High
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.