Database/AI/ML frameworks & serving
NVIDIA NemoClaw: insufficiently protected credentials allow information disclosure and data tampering
Impact
NVIDIA states that NemoClaw insufficiently protects credentials, and that a successful exploit might lead to information disclosure and data tampering. Credentials handled by AI tooling on a GPU fleet are usually the keys to model registries, object storage or downstream API endpoints, so exposure tends to be worth more than the host it was read from. The record gives no detail on where the credentials are stored or which principal can read them - CVSS scores it local, low-privilege, requiring user interaction, with high confidentiality and low integrity impact (5.6). Do not assume the exposure is limited to the local machine until the advisory says so.
Who can reach it
A local, low-privileged user on the host running NemoClaw, plus user interaction (CVSS AV:L/AC:L/PR:L/UI:R). Some authenticated local access is required.
What to do
Update NemoClaw to the fixed release listed in NVIDIA advisory bundle 5872; the record does not name a fixed version, so take it from the advisory. Because the issue concerns exposed credentials, rotate any credentials NemoClaw held after patching - patching alone does not undo an exposure that already happened.
References
Related entries
- Linux perf/x86/amd/uncore - memory leak in the events array: Per-CPU northbridge and last-level-cache uncore contextsCVE-2022-49784 · Linux perf/x86/amd/uncore - memory leak in the events arrayMedium
- PyTorch (flatbuffer loader): Out-of-bounds read parsing flatbuffer modelCVE-2024-31584 · PyTorch (flatbuffer loader)Medium
- vLLM: crafted request to the Gemma4 unified parser crashes the inference serverCVE-2026-103241 · vLLM (Gemma4UnifiedParser, rust/src/parser/src/unified/gemma4.rs)Medium
- Ollama: model-name validation flaw in the Pull API allows unauthenticated server-side request forgeryCVE-2026-108836 · Ollama (Pull API model-name validation, types/model/name.go)Medium
- Keras (HDF5 ExternalLink, incomplete fix): Arbitrary HDF5 file readCVE-2026-12480 · Keras (HDF5 ExternalLink, incomplete fix)Medium
- Keras: unvalidated dataset sizes in .keras loading let a poisoned model exhaust node memoryCVE-2026-12570 · Keras (.keras model loading, H5IOStore.__getitem__)Medium
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.