Database/AI/ML frameworks & serving
NVIDIA NeMo Speech: RCE and privilege escalation in the speech data explorer
Impact
The speech data explorer component executes attacker-controlled data, yielding code execution and escalation of privileges on the host running it, along with information disclosure and data tampering. The explorer is a dataset-inspection tool that people point at corpora they did not produce, which is exactly the trust boundary this breaks: reviewing a tenant's or a third party's dataset hands that dataset's author code on the box. NVIDIA also lists privilege escalation for this id, unlike the other NeMo Speech issues in the same bulletin, so a compromise here can reach beyond the analyst's own account on a shared workstation or utility node.
Who can reach it
Anyone who can place a dataset or manifest that the speech data explorer is then used to inspect. No prior authentication to NeMo; user interaction required, local vector (AV:L/UI:R).
What to do
Update NeMo Speech per NVIDIA security bulletin 2026/5885 and restart the data explorer. The record does not state a fixed version - check the bulletin. Meanwhile do not run the explorer against untrusted datasets on a host with cluster or registry credentials; run it as an unprivileged user in a throwaway container.
References
Related entries
- BentoML (`safe_extract_tarfile`): Tar extraction escape despite the "safe" helperCVE-2026-27905 · BentoML (`safe_extract_tarfile`)High
- llama.cpp (`ggml_nbytes`): Integer overflow in the core ggml size calculationCVE-2026-33298 · llama.cpp (`ggml_nbytes`)High
- BentoML (`docker.system_packages`): Command injection through the package list fieldCVE-2026-33744 · BentoML (`docker.system_packages`)High
- BentoML (cloud deployment path, setup.sh generation in deployment.py): The March fix that added shlex.quote to theCVE-2026-35043 · BentoML (cloud deployment path, setup.sh generation in deployment.py)High
- SGLang (`replay_request_dump.py`): Insecure `pickle.load()` on a `.pkl` dumpCVE-2026-3989 · SGLang (`replay_request_dump.py`)High
- llama.cpp (`llama_batch_init`): Integer overflow from unchecked multiplicationCVE-2026-43627 · llama.cpp (`llama_batch_init`)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.