GPU VulnDB

Database/AI/ML frameworks & serving

NVIDIA DALI: out-of-bounds reads while decoding pipeline input lead to disclosure and crashes

CVSS 7.3CVE-2026-65147AI/ML frameworks & serving+1 more CVEscurated

Impact

DALI is the data loading and augmentation library that feeds training and inference pipelines on GPU nodes, so it parses whatever media or serialized input a job is pointed at. Two ids in NVIDIA bulletin 2026/5900 (CVE-2026-65147 and CVE-2026-65150) describe uncontrolled out-of-bounds reads with the same score and the same fix; NVIDIA states the consequences as information disclosure, data tampering and denial of service. On a shared cluster the practical effect is that a job processing attacker-supplied data can read memory it should not and kill the pipeline process; the record does not claim a container or host escape. Credibly reaching it requires getting a crafted input into someone's pipeline.

Who can reach it

Local, authenticated low-privileged user, with user interaction (CVSS AV:L/AC:L/PR:L/UI:R) - in practice a tenant or dataset contributor who can get a crafted file into a DALI pipeline that another user runs.

What to do

Upgrade the DALI package to the fixed version named in NVIDIA bulletin 2026/5900 and rebuild the training/inference images that pin it, then restart the affected jobs or serving processes. No node reboot or firmware work is involved. The NVD record does not carry the fixed version string - read the bulletin before scheduling.

Also covers 1 CVE

The vendor assigned a separate id to each affected code path. They share this advisory, this score and this fix, so they are one entry here.

CVE-2026-65150

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.