Ddos 
NVIDIA has released a security update for its TensorRT-LLM Framework, addressing a high-severity vulnerability that could expose users to serious security risks, including remote code execution, data tampering, and information disclosure.
The vulnerability, identified as CVE-2025-23245 and assigned a CVSS base score of 8.8 (High), affects the Python executor component of the TensorRT-LLM Framework. According to NVIDIA’s bulletin, the flaw stems from insecure handling in the socket-based Inter-Process Communication (IPC) system, specifically related to Python’s pickle serialization/deserialization mechanism.
“NVIDIA TensorRT-LLM for any platform contains a vulnerability in Python executor where an attacker may cause a data validation issue by local access to the TRTLLM server. A successful exploit of this vulnerability may lead to code execution, information disclosure and data tampering,” the company warned.
To mitigate the issue, NVIDIA has implemented HMAC encryption by default in both the main and release branches of TensorRT-LLM.
“TensorRT-LLM added a security enhancement for its socket-based Inter-Process Communication (IPC)… HMAC encryption will be set to encryption ON in both release branch and main branch by default,” the bulletin reads.
Users are strongly discouraged from disabling this setting, as doing so will reintroduce the vulnerability: “Warning: Disabling this feature will make you vulnerable to the security issue.”
If necessary, the encryption feature can be manually disabled by setting use_hmac_encryption = False in the ZeroMqQueue class located in:
- tensorrt_llm/executor/ipc.py for the main branch
- tensorrt_llm/executor.py for release 0.18
Affected platforms include Windows, Linux, and macOS, with all versions prior to 0.18.2 being vulnerable. Users are advised to:
- Update to TensorRT-LLM v0.18.2 or later
- Clone the latest version from GitHub Main or v0.18.2 release tag