Executive brief
NVIDIA TRT-LLM, a library used to accelerate and optimize Large Language Model (LLM) performance, contains a software flaw. An attacker could exploit this issue to cause the software to crash, leading to a denial of service. This could disrupt AI-driven operations and services that rely on this library for processing.
Technical details
A vulnerability exists in NVIDIA TRT-LLM (all platforms) classified as CWE-690 (Unchecked Return Value to NULL Pointer Dereference). The issue occurs when the software fails to verify the return value of a function before using it as a pointer, which can lead to a null pointer dereference. This is a local attack vector requiring user interaction (CVSS:3.1/AV:L/AC:L/PR:N/UI:R). A successful exploit allows an attacker to trigger a crash of the application, resulting in a denial of service (DoS). No impact on confidentiality or integrity was reported.
Affected products
- NVIDIA TRT-LLM All platforms
Timeline
- 2026-05-20: disclosed: Initial publication of CVE-2026-24160