Database/AI/ML frameworks & serving
NVIDIA Transformer Engine: unsafe deserialization of untrusted data allows code execution
Impact
Transformer Engine can deserialize untrusted data when it loads it. NVIDIA says a successful exploit may lead to code execution, information disclosure and denial of service. Transformer Engine runs inside training and inference jobs on GPU nodes, so the code would run with the job's privileges and could reach its GPUs, its mounted datasets and checkpoints, and any credentials in the job environment. The record does not name the affected code path.
Who can reach it
Local. No privileges are needed, but a user must be tricked into loading attacker-supplied data, such as a checkpoint or other serialized file, with Transformer Engine (CVSS AV:L/PR:N/UI:R).
What to do
Upgrade Transformer Engine to the fixed release named in NVIDIA bulletin 5927. The record does not give the fixed version. It is a library inside job images and Python environments, so rebuild the affected containers or environments and restart the jobs. No node reboot is needed. Until then, load checkpoints and serialized artifacts only from trusted sources.
References
Related entries
- NVIDIA Megatron-LM: malicious checkpoint files execute code when loadedCVE-2026-65110 · NVIDIA Megatron-LM (checkpoint loading)High
- NVIDIA NeMo Speech: code injection from malicious input, no user interaction neededCVE-2026-65111 · NVIDIA NeMo SpeechHigh
- NVIDIA Megatron-LM: unsafe deserialization when loading checkpoints allows code executionCVE-2026-65132 · NVIDIA Megatron-LM (checkpoint loading)High
- NVIDIA Isaac Lab: eval injection allows code execution from attacker-influenced inputCVE-2026-65141 · NVIDIA Isaac LabHigh
- NVIDIA Model-Optimizer: deserialization of untrusted data leads to code executionCVE-2026-65142 · NVIDIA Model-Optimizer (model/checkpoint deserialization)High
- NVIDIA NeMo: a crafted model_config.yaml injects unsafe parameters into dataset loadingCVE-2026-65178 · NVIDIA NeMo (dataset-loading workflow, model_config.yaml parameter injection)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.