Executive brief
NVIDIA TensorRT-LLM is a software library used to optimize the performance of large language models on Linux systems. A security vulnerability has been identified that could allow an attacker to interfere with how the software generates code. If successfully exploited, this could lead to unauthorized code execution, the theft of sensitive information, or the tampering of data within the AI environment.
Technical details
NVIDIA TensorRT-LLM for Linux is vulnerable to improper control of code generation (CWE-829). The flaw exists in the way the library handles functionality from an untrusted control sphere. Exploitation requires local access with high privileges and involves high complexity, including a requirement for user interaction. A successful attack allows for arbitrary code execution, data tampering, and unauthorized information disclosure within the context of the affected application. The vulnerability affects versions up to and including v1.3.0 rc12.
Affected products
- NVIDIA TensorRT-LLM Up to and including v1.3.0 rc12
Timeline
- 2026-07-14: disclosed: Initial publication of the CVE record.
- 2026-07-14: advisory: NVIDIA released advisory information via NVD.