NVIDIA Triton Inference Server processes HTTP requests and, due to an unchecked input condition, can be forced to crash by a specially crafted request header. This flaw leads to a complete disruption of the inference service for all clients that rely on the server, causing a denial of service.

The vulnerability applies to NVIDIA’s Triton Inference Server. The affected versions are not enumerated in the advisory, so any deployment that has not received a corrective update from NVIDIA may be susceptible.

The CVE has a CVSS score of 7.5, indicating a high risk level. No EPSS data is available, and the issue is not listed in CISA’s KEV catalog. Attackers are likely to exploit the weakness by sending a malformed header over the network, a straightforward technique that requires only the ability to reach the server’s interface.