NVIDIA Triton Inference Server is vulnerable to an internal state corruption flaw that can trigger a denial of service. The weakness is a race condition impacting the server’s ability to process inference requests correctly. When exploited, the server can become unresponsive, leading to service disruption for any application relying on the inference engine. The Primary impact is on availability, as reflected by a CVSS score of 7.5.

All deployments of NVIDIA Triton Inference Server are affected. No specific firmware or software release numbers were listed in the available data, so any version that contains the vulnerable code path requires assessment of the latest builds from NVIDIA to determine if the issue has been fixed.

The assessed criticality is high, with a CVSS score of 7.5, and the EPSS score is reported as less than 1%, indicating a low current likelihood of exploitation. This vulnerability is not recorded in CISA’s Known Exploited Vulnerabilities catalog. The description does not detail how the race condition is triggered, but the likely attack vector is through the network interface by sending specially crafted inference requests, as is common with Triton server vulnerabilities. This inference is based on the nature of the software and typical exploits of internal state corruption in similar systems.