GPU VulnDB

Database/AI/ML frameworks & serving

NVIDIA NemoClaw for Linux: installer downloads code without an integrity check

CVE-2026-65097AI/ML frameworks & servingcurated

Impact

NVIDIA reports that NemoClaw's Linux installation scripts fetch code without verifying its integrity. An attacker positioned to tamper with what the installer downloads can get their code executed in the context of the install, which NVIDIA says may lead to code execution, privilege escalation, information disclosure and data tampering. The CVSS vector requires user interaction and rates attack complexity high, which fits a tamper-the-download scenario rather than a directly triggerable remote bug. The exposure on a fleet is at provisioning time: if this installer runs as part of node or image build, a compromised fetch is baked into every node built from that image. The advisory record is thin and does not name affected or fixed versions - treat the details above as the limit of what is established.

Who can reach it

Someone able to interfere with the code the installer downloads - control of the network path, a hijacked or spoofed download endpoint, or a compromised mirror. Requires an operator to actually run the installation script; no pre-existing access to the target host is needed.

What to do

Consult NVIDIA's advisory (product-security entry 2026/5872) for the affected and fixed versions - the record given here names neither, so do not assume a version is safe. Until then, run NemoClaw installs only over paths you control, verify downloaded artefacts against NVIDIA-published checksums or signatures out of band, and do not run the installer as root on nodes that carry fleet credentials. If it is already in an image build pipeline, that pipeline is the thing to fix first.

References

Related entries

All AI/ML frameworks & serving entries

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.