Critical Deserialization Vulnerability (CVE-2025-23254) in NVIDIA TensorRT-LLM: Upgrade to Secure Your Systems Now

Nox90 Engineering
May 4
2 min read

Image for post about Detailed Report on CVE-2025-23254: Deserialization Vulnerability in NVIDIA TensorRT-LLM

Overview:

CVE-2025-23254 is a high-severity vulnerability identified in NVIDIA's TensorRT-LLM framework. This security flaw primarily affects the Python executor component, specifically within its socket-based Inter-Process Communication (IPC) system that utilizes Python's pickle serialization and deserialization. The vulnerability is categorized under CWE-502 (Deserialization of Untrusted Data), which allows attackers to execute arbitrary code, access sensitive information, and manipulate data without proper validation.

Affected Versions:

All versions of NVIDIA TensorRT-LLM prior to v0.18.2 across all platforms, including Windows, Linux, and macOS.

Impact:

Successful exploitation of this vulnerability can lead to: - Remote Code Execution - Information Disclosure - Data Tampering

CVSS Score:

Base Score: 8.8 (High)
Vector: CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H

Exploitation Details:

The vulnerability requires local access to the TRTLLM server, where an attacker can exploit improper data validation during the deserialization process. This flaw enables the execution of arbitrary code, thereby compromising the system's integrity and confidentiality.

Mitigation Strategies:

Upgrade Immediately: Users are strongly advised to update to TensorRT-LLM v0.18.2 or later. The updated version addresses the vulnerability by introducing security enhancements.
Enable HMAC Encryption: NVIDIA has implemented HMAC (Hash-based Message Authentication Code) encryption by default for all socket-based IPC operations. Users should ensure this feature remains enabled to maintain the integrity and authenticity of serialized data. Disabling it will reintroduce the vulnerability.
Monitoring and Alerts: Regularly check NVIDIA's security bulletins and subscribe to their notifications for timely updates and patches.

Acknowledgements:

NVIDIA credits Avi Lumelsky of Oligo Security for reporting this vulnerability.

Conclusion:

CVE-2025-23254 highlights the critical importance of secure deserialization practices in software development. Organizations using NVIDIA TensorRT-LLM should prioritize implementing the recommended updates and security configurations to mitigate potential risks.