CVE-2023-31031 is a medium severity flaw (CVSS 4.2) in NVIDIA DGX A100 firmware. A local user may cause a heap based buffer overflow in the SBIOS. NVIDIA's fix is SBIOS 1.25 or later, delivered in the firmware update container 23.12.1.
What's affected
NVIDIA's bulletin (revision 1.2, 25 January 2024) lists these fixed versions, delivered through the NVIDIA DGX A100 Firmware Update Container version 23.12.1 on the NVIDIA Enterprise Support Portal.
- BMC: all versions prior to 00.22.05 are affected. Fixed: 00.22.05.
- SBIOS: all versions prior to 1.25 are affected. Fixed: 1.25.
The NVD record names DGX Station A100 and DGX Station A800. NVIDIA's table lists DGX A100 as the product for SBIOS fixes, so check with NVIDIA if you run a Station or A800 model.
How to check if you're exposed
Read the SBIOS version from the host:
sudo dmidecode -s bios-version
The fixed SBIOS is 1.25 or later. This is a device or appliance issue, not a package you pull from a registry, so a dependency scan will not see it. Safeguard's free in-browser scan at https://gold.safeguard.sh/scan reads dependency manifests and will not detect firmware running on hardware. Use your asset inventory instead.
What to do now
- Find every NVIDIA DGX A100 system and record its BMC and SBIOS versions.
- Update to SBIOS 1.25 or later by installing the DGX A100 Firmware Update Container 23.12.1 from the NVIDIA Enterprise Support Portal.
- Keep BMC interfaces on a separate management network that only administrators can reach.
Why it matters
The record lists CWE-122, heap based buffer overflow, and CWE-787, out of bounds write. The vector needs local access and high privileges, so the score is only 4.2. NVIDIA says the impact could still be code execution, denial of service, information disclosure and data tampering.
Firmware bugs rated Medium are still worth including in a normal maintenance window, since the update container carries many fixes at once. The CVE is not on the CISA KEV list.
On a workstation class system such as the DGX Station, the person using the machine often has local administrator rights and runs the update container. That makes the local access requirement easier to meet than on a locked down data centre rack. Confirm who has shell access to each system and whether the system model really appears in NVIDIA's table. If you cannot confirm that your model is covered, ask NVIDIA support before you assume the system is safe, and keep the question in the ticket until it is answered.
Quick check
Why is the score low?
The vector requires local access and high privileges.
Which systems?
The record names DGX Station A100 and A800. NVIDIA's table names DGX A100.
Sources
Last reviewed: 19 October 2026.