CVE-2024-0088 is a Medium severity (CVSS 5.5) flaw in the shared memory APIs of NVIDIA Triton Inference Server for Linux, where a user can cause an improper memory access through a network API. Versions 20.10 through 24.03 are affected. Upgrade to 24.04 or later and protect the shared memory APIs.
What's affected
- Product: NVIDIA Triton Inference Server for Linux.
- Affected versions: 20.10 through 24.03, per the NVIDIA bulletin. This range is wider than CVE-2024-0087, which starts at 22.09.
- Fixed version: 24.04.
- Severity: Medium, CVSS 5.5, network vector, high attack complexity, high privileges required. Possible impacts per the record are denial of service and data tampering. The record and the bulletin agree.
- NVIDIA's mitigation: upgrade to 24.04 or later, and ensure that logging and shared memory APIs are protected for use by authorized users.
- Exploitation: the record's KEV flag is false, and the bulletin does not report exploitation.
How to check if you're exposed
Find your Triton releases. Where Triton runs from a container, inspect image tags:
docker images | grep -i tritonserver
Anything from 20.10 through 24.03 is in range. Image scanning is a reliable way to find old Triton builds that nobody remembers deploying. The free in-browser dependency scan covers manifests, and the container image scan demo below shows the approach for images.
Then check who can call the shared memory APIs, since the bulletin asks that they be restricted to authorized users.
What to do now
- Inventory Triton deployments and their versions, including old test environments.
- Upgrade versions 20.10 through 24.03 to 24.04 or later.
- Restrict shared memory and logging APIs to authorized users.
- Apply it together with CVE-2024-0087, since the fixed version is the same.
- Rescan images after the rollout to confirm no old tags remain.
Why it matters
The record maps this to CWE-119 (CWE-119) and CWE-787 (CWE-787): improper restriction of operations within the bounds of a memory buffer, and out of bounds write. In plain terms, a request through the shared memory API makes the server touch memory it should not.
The Medium score comes from the required privileges and the high complexity. Yet the same bulletin fixes a Critical flaw in the same product, so the practical advice is simple. Move to 24.04 and close both issues in one upgrade.
Teams also forget old test and staging servers. The affected range reaches back to 20.10, so a Triton instance stood up for a proof of concept years ago can still be reachable inside your network. Look for these first. They rarely get upgraded, they often run with generous network access, and they are the easiest place for an unnoticed memory access flaw to sit.
Try it yourself
Click through it at your own pace, or open it full screen.
Quick check
Which Triton versions are affected by CVE-2024-0088?
20.10 through 24.03, according to the NVIDIA bulletin. The fix is in 24.04.
Can I patch CVE-2024-0087 and CVE-2024-0088 together?
Yes. Both list 24.04 as the fixed version, so one upgrade covers both.
Sources
- NVIDIA security bulletin 5535 (Triton Inference Server, May 2024)
- Safeguard Gold entry for CVE-2024-0088
Last reviewed: 19 October 2026.
