EUVD-2025-208976
GHSA-gf35-gc84-4gf8
CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
Lifecycle Timeline
3Description
NVIDIA Triton Inference Server contains a vulnerability where an attacker may cause internal state corruption. A successful exploit of this vulnerability may lead to a denial of service.
Analysis
NVIDIA Triton Inference Server contains a race condition vulnerability (CWE-362) that allows unauthenticated remote attackers to corrupt internal server state, resulting in a denial of service. The vulnerability affects NVIDIA Triton Inference Server across multiple versions and can be exploited over the network with low attack complexity requiring no privileges or user interaction. With a CVSS score of 7.5 (High) and an EPSS score not provided, this represents a significant availability risk for organizations running AI/ML inference workloads.
Technical Context
NVIDIA Triton Inference Server is an open-source inference serving software that enables deployment of trained AI models from multiple frameworks (TensorFlow, PyTorch, ONNX, etc.) in production environments. The vulnerability stems from CWE-362 (Concurrent Execution using Shared Resource with Improper Synchronization, commonly known as a Race Condition), where multiple threads or processes access shared resources without proper locking mechanisms. In the context of Triton, this likely involves race conditions during request handling, model loading/unloading, or shared memory operations that manage inference state. The affected product is identified via CPE as cpe:2.3:a:nvidia:triton_inference_server:*:*:*:*:*:*:*:*, indicating broad version coverage pending specific version details from the vendor advisory.
Affected Products
NVIDIA Triton Inference Server is affected across multiple versions as indicated by the CPE identifier cpe:2.3:a:nvidia:triton_inference_server:*:*:*:*:*:*:*:*. The wildcard version field suggests broad impact, though specific version ranges requiring remediation are detailed in NVIDIA's security bulletin available at https://nvidia.custhelp.com/app/answers/detail/a_id/5790. Organizations should consult this advisory to determine if their deployed versions fall within the vulnerable range and require immediate patching.
Remediation
Organizations should immediately consult the NVIDIA security bulletin at https://nvidia.custhelp.com/app/answers/detail/a_id/5790 to identify affected versions and obtain the patched release of NVIDIA Triton Inference Server. Apply the vendor-provided security update as soon as possible following standard change management procedures. Until patching can be completed, implement network-level mitigations including restricting access to Triton Server endpoints to trusted IP ranges using firewall rules or network policies, deploying rate limiting to reduce race condition exploitation windows, and monitoring for unusual patterns of concurrent requests or server crashes that may indicate exploitation attempts. Consider placing Triton behind a reverse proxy or API gateway to add additional request filtering and anomaly detection capabilities.
Priority Score
Share
External POC / Exploit Code
Leaving vuln.today