August 4, 2025
5 min read
Christopher Tepedino
Nvidia releases urgent update to fix critical security flaws in its Triton AI inference server, preventing potential AI model takeovers.
Nvidia Patches Critical Vulnerabilities in Triton AI Inference Server
Technology giant Nvidia released a software update on Saturday to address critical security vulnerabilities in its Triton inference server, widely used for deploying artificial intelligence (AI) models. Triton is Nvidia’s open-source inference server designed to optimize AI model deployment. Recently, cybersecurity firm Wiz identified a chain of vulnerabilities in Triton that could allow attackers to take full control of AI servers, potentially leading to data theft and manipulation of AI responses. Wiz’s head of vulnerability research, Nir Ohfeld, explained to Cointelegraph that the attack begins with a minor bug causing the server to leak a small piece of secret internal data. This leaked data can then be exploited to trick legitimate server features into granting control over private system components. Once an attacker gains this foothold, they can escalate privileges and achieve complete server takeover. While the exact number of Triton users is unknown, major enterprises such as Microsoft, Amazon, Oracle, Siemens, and American Express have been reported to use the platform. According to a 2021 Nvidia press release, over 25,000 companies deploy Nvidia’s AI stack worldwide. The vulnerabilities have been assigned CVE identifiers CVE-2025-23319, CVE-2025-23320, and CVE-2025-23334. Nvidia has urged all users to update to Triton Inference Server version 25.07 or newer to fully mitigate the risks. Ohfeld emphasized the importance of applying the patch immediately, stating, "The single most important step is to update to the patched version of the Nvidia Triton Inference Server (version 25.07 or newer). This directly fixes the entire vulnerability chain." As of now, there is no evidence these vulnerabilities have been exploited in the wild, but given Triton's popularity in AI workloads, the risk remains significant.Security Challenges in Emerging Technologies
Security vulnerabilities continue to challenge emerging technologies in 2025. In the crypto sector, exploits related to access control flaws and smart contract bugs have resulted in losses exceeding $3.1 billion in the first half of the year, surpassing the total losses of 2024, according to blockchain security auditor Hacken. Experts also warn that AI agents and quantum computing technologies may introduce new cyber threats in the near future, underscoring the need for robust security measures across all advanced tech domains.Source: Nvidia releases update for 'critical' vulnerabilities in AI stack