Executive brief
NVIDIA Triton Inference Server, a platform used to deploy and manage AI models in production, is vulnerable to a flaw that can cause the service to crash. An attacker can exploit this issue remotely without needing any special permissions or user interaction. If successful, this would result in a denial of service, preventing legitimate users and applications from accessing AI inference capabilities.
Technical details
A vulnerability exists in NVIDIA Triton Inference Server for Linux (CWE-910) involving the use of an expired file descriptor. The flaw can be triggered by a remote, unauthenticated attacker over the network with low attack complexity. Successful exploitation leads to a denial of service (DoS) condition by causing the server to reference a file descriptor that is no longer valid or has been closed. The vulnerability affects versions up to and including 26.04.
Affected products
- NVIDIA Triton Inference Server versions up to and including 26.04
Timeline
- 2026-07-14: disclosed: Initial publication of the CVE record.