High severity NVIDIA DCGM Exporter vulnerability threatens internet exposed GPU servers

Unauthenticated attackers can exploit a high-severity vulnerability in NVIDIA DCGM Exporter to crash GPU monitoring services and disrupt artificial intelligence workloads. Help Net Security, CSO Online, and The Register all report that the security flaw allows remote attackers to cause denial-of-service conditions on exposed infrastructure. According to CSO Online, the component is also vulnerable to information disclosure attacks. [1][2][3]

The vulnerability is tracked as CVE-2026-47483 and carries a CVSS score of 8.2. Help Net Security reported that cybersecurity company Lava discovered the flaw and reported it to NVIDIA. NVIDIA published an official security bulletin regarding the issue on July 28, 2026. [1]

The issue stems from an unauthenticated resource exhaustion flaw in the software. CSO Online reported that thousands of exposed DCGM Exporter instances leak GPU telemetry, and some expose profiling endpoints that allow remote attackers to crash the service. Help Net Security noted that hundreds of internet-exposed graphics processing unit servers were open to the vulnerability. [1][2]

GPU servers are specialized computers built around graphics processing units to run artificial intelligence, machine learning, and scientific computing workloads. CSO Online explained that enterprises use the NVIDIA DCGM Exporter software component to monitor their artificial intelligence training and inference infrastructure. [1][2]